CN110363788A - A kind of video object track extraction method and device - Google Patents
A kind of video object track extraction method and device Download PDFInfo
- Publication number
- CN110363788A CN110363788A CN201910505008.6A CN201910505008A CN110363788A CN 110363788 A CN110363788 A CN 110363788A CN 201910505008 A CN201910505008 A CN 201910505008A CN 110363788 A CN110363788 A CN 110363788A
- Authority
- CN
- China
- Prior art keywords
- image
- pixel
- transparency
- mask
- transparency mask
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/174—Facial expression recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20036—Morphological image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The disclosure discloses a kind of video object track extraction method and device, transparency estimated value is recalculated to obtain the first transparency mask of the first image by measuring the confidence level of prospect background pixel pair first, then new picture is generated by superposition grayscale information and obtain the second transparency mask of the first image, and further correct the first transparency mask of the first image, finally the foreground target of frame image a certain in video is extracted using the modified first transparency mask, and specified target is retrieved in all foreground targets, then according to the time sequencing of all frames, the track of the specified target is generated according to all images for including the specified target.The disclosure can comprehensively utilize the confidence level and grayscale information of prospect background pixel pair in a certain frame image in video, provide a kind of scheme of new video object trajectory extraction.
Description
Technical field
The disclosure belongs to field of image processing, in particular to a kind of video object track extraction method and device.
Background technique
In safety-security area, there is the demand of many trajectory extractions to set the goal for video middle finger.
However, in the prior art, although there are the schemes of enough extraction video object tracks, on how to benefit
With prospect background pixel to and grayscale information go forward side by side onestep extraction target trajectory to extract video foreground target, there has been no related new
The implementation method of grain husk.
Summary of the invention
Present disclose provides a kind of video object track extraction methods, include the following steps:
S100 divides all foreground pixel set F, the had powerful connections picture in the image for the first image in video
Element set B and all unknown pixel set Z；Wherein, the first image is a certain frame image extracted from the video；
S200 gives certain prospect background pixels to (Fi, Bj), each unknown pixel Z is measured according to the following formulakIt is saturating
Lightness
Wherein, IkFor unknown pixel ZkRGB color value, the foreground pixel FiFor apart from unknown pixel ZkNearest m
Foreground pixel, the background pixel BjAlso for apart from unknown pixel ZkM nearest background pixel, the prospect background pixel pair
(Fi, Bj) amount to m2Group；
S300, for the m2Each group of prospect background pixel in group is to (Fi, Bj) and its it is correspondingAccording to as follows
Formula measures prospect background pixel to (Fi, Bj) confidence level nij:
Wherein, σ value 0.1, and choose the highest MAX (n of confidence levelij) corresponding to that group prospect background pixel to for
(FiMAX, BjMAX)；
S400 calculates each unknown pixel Z according to the following formulakTransparency estimated value
S500, according to each unknown pixel ZkTransparency estimated valuePrimarily determine the first of the first image
Transparency mask；
S600, to the first image superposition grayscale information to generate the second image, and it is all to divide it to second image
Foreground pixel set, all background pixel set and all unknown pixel set；
S700 executes step S200 to S500 for second image, to determine that the first transparency of the second image hides
Cover, and using the first transparency mask of second image as the second transparency mask of the first image；
S800, using the second transparency mask of the first image, the first transparency for correcting the first image is hidden
Cover；
S900 corrects the first transparency mask of resulting first image according to step S800, to the first of the video
Foreground target in image extracts, and specified target is retrieved in all foreground targets, then according to the time of all frames
Sequentially, the track of the specified target is generated according to all images for including the specified target.
In addition, the disclosure further discloses a kind of video object trajectory extraction device, comprising:
First division module, is used for: for the first image in video, dividing all foreground pixel set in the image
F, all background pixel set B and all unknown pixel set Z；Wherein, the first image is extracted from the video
A certain frame image；
First metric module, is used for: giving certain prospect background pixels to (Fi, Bj), measurement is each not according to the following formula
Know pixel ZkTransparency
Wherein, IkFor unknown pixel ZkRGB color value, the foreground pixel FiFor apart from unknown pixel ZkNearest m
Foreground pixel, the background pixel BjAlso for apart from unknown pixel ZkM nearest background pixel, the prospect background pixel pair
(Fi, Bj) amount to m2Group；
Second metric module, is used for: for the m2Each group of prospect background pixel in group is to (Fi, Bj) and its it is corresponding
'sMeasurement prospect background pixel is to (F according to the following formulai, Bj) confidence level nij:
Wherein, σ value 0.1, and choose the highest MAX (n of confidence levelij) corresponding to that group prospect background pixel to for
(FiMAX, BjMAX)；
Computing module is used for: calculating each unknown pixel Z according to the following formulakTransparency estimated value
Determining module is used for: according to each unknown pixel ZkTransparency estimated valuePrimarily determine described first
First transparency mask of image；
Second division module, is used for: to the first image superposition grayscale information to generate the second image, and to second figure
As dividing its all foreground pixel set, all background pixel set and all unknown pixel set；
Calling module again is used for: being directed to second image, is called first metric module, the second measurement again
Module, computing module and determining module, to determine the first transparency mask of the second image, and by the first of second image
Second transparency mask of the transparency mask as the first image；
Correction module is used for: using the second transparency mask of the first image, correcting the first of the first image
Transparency mask；
Extraction module is used for: according to the first transparency mask of resulting first image of correction module, to the video
Foreground target in first image extracts, and specified target is retrieved in all foreground targets, then according to all frames
Time sequencing generates the track of the specified target according to all images for including the specified target.
By the method and device, the disclosure is capable of the confidence level and gray scale letter of Prospects of Comprehensive Utilization background pixel pair
Breath provides a kind of scheme of new video object trajectory extraction.
Detailed description of the invention
Fig. 1 is the schematic diagram of one embodiment the method in the disclosure；
Fig. 2 is the schematic diagram of another embodiment described device in the disclosure.
Specific embodiment
In order to make those skilled in the art understand that technical solution disclosed by the disclosure, below in conjunction with embodiment and related
The technical solution of each embodiment is described in attached drawing, and described embodiment is a part of this disclosure embodiment, without
It is whole embodiments.Term " first " used by the disclosure, " second " etc. rather than are used for for distinguishing different objects
Particular order is described.In addition, " comprising " and " having " and their any deformation, it is intended that covering and non-exclusive packet
Contain.Such as contain the process of a series of steps or units or method or system or product or equipment are not limited to arrange
Out the step of or unit, but optionally further include the steps that not listing or unit, or further includes optionally for these mistakes
Other intrinsic step or units of journey, method, system, product or equipment.
Referenced herein " embodiment " is it is meant that a particular feature, structure, or characteristic described can wrap in conjunction with the embodiments
It is contained at least one embodiment of the disclosure.Each position in the description occur the phrase might not each mean it is identical
Embodiment, nor the independent or alternative embodiment with other embodiments mutual exclusion.It will be appreciated by those skilled in the art that
, embodiment described herein can combine with other embodiments.
Show referring to the process that Fig. 1, Fig. 1 are a kind of video object track extraction methods that one embodiment provides in the disclosure
It is intended to.As shown, described method includes following steps:
S100 divides all foreground pixel set F, the had powerful connections picture in the image for the first image in video
Element set B and all unknown pixel set Z；Wherein, the first image is a certain frame image extracted from the video；
It is understood that there are many means for dividing foreground pixel, background pixel and unknown pixel to image, can be artificial
Mark, can also by way of machine learning or data-driven, can also be according to corresponding prospect threshold value, background threshold come
Mark off all foreground and background pixels and its corresponding set；Once foreground and background pixel is divided out, unknown pixel,
It corresponds to set also with regard to being divided out naturally；
In addition, the first image may is that when video is played when extracting to video foreground target, ring
Current video broadcasting should be suspended, and carry out the interception of present frame immediately to pause picture, in the operation of user to obtain
Obtain the first image；The first image is also possible to: when video is there is no being played, in response to the operation of user, choosing at random
The a certain frame or a few frames in video are selected, using wherein a certain frame image as the first image.In any case, it is to be understood that the party
The foreground target that method can be used for each frame image in video extracts.Preferably, the first image is first frame in video
Image.
S200 gives certain prospect background pixels to (Fi, Bj), each unknown pixel Z is measured according to the following formulakIt is saturating
Lightness
Wherein, IkFor unknown pixel ZkRGB color value, the foreground pixel FiFor apart from unknown pixel ZkNearest m
Foreground pixel, the background pixel BjAlso for apart from unknown pixel ZkM nearest background pixel, the prospect background pixel pair
(Fi, Bj) amount to m2Group；
To those skilled in the art, theoretically, the selection of m can make corresponding prospect background pixel to being
Part sample, can also be with exhaustive whole image；For step S200, it is intended to the color and prospect background by unknown pixel
The color relationship of pixel pair estimates the transparency of unknown pixel；In addition, the selection of m can also further combined with neighborhood territory pixel with
Between unknown pixel color, texture, gray scale, brightness, in terms of feature；
S300, for the m2Each group of prospect background pixel in group is to (Fi, Bj) and its it is correspondingAccording to as follows
Formula measures prospect background pixel to (Fi, Bj) confidence level nij:
Wherein, σ value 0.1, and choose the highest MAX (n of confidence levelij) corresponding to that group prospect background pixel to for
(FiMAX, BjMAX)；
It is understood that the value of σ is empirical value or statistical value or simulation value, step S300 is further screened using confidence level
Prospect background pixel pair, and for subsequent step by the prospect background pixel further screened to estimating that unknown pixel is transparent
Degree；
S400 calculates each unknown pixel Z according to the following formulakTransparency estimated value
S500, according to each unknown pixel ZkTransparency estimated valuePrimarily determine the first of the first image
Transparency mask；
This is to say, after the transparency estimated value of each unknown pixel obtains, the present embodiment with regard to primarily determining naturally
First transparency mask of the first image；Why say be naturally, be because transparency mask can be considered as by
By selected those respective pixels composition of certain value (or value range)；
S600, to the first image superposition grayscale information to generate the second image, and it is all to divide it to second image
Foreground pixel set, all background pixel set and all unknown pixel set；
For the step, the present embodiment is contemplated that gray scale is believed in view of each pixel is in addition to the effect of RGB color
Cease the influence to pixel；Therefore, after being superimposed grayscale information, transparency mask is modified using following steps.
S700 executes step S200 to S500 for second image, to determine that the first transparency of the second image hides
Cover, and using the first transparency mask of second image as the second transparency mask of the first image；
S800, using the second transparency mask of the first image, the first transparency for correcting the first image is hidden
Cover；
S900 corrects the first transparency mask of resulting first image according to step S800, to the first of the video
Foreground target in image extracts, and specified target is retrieved in all foreground targets, then according to the time of all frames
Sequentially, the track of the specified target is generated according to all images for including the specified target.
So far, the confidence level and grayscale information of disclosure Prospects of Comprehensive Utilization background pixel pair provides a kind of new video
The scheme that target trajectory extracts.It is understood that in this scenario, the extraction of video foreground target therein is one and infinitely forces
Close process, due to the transition of color, gray scale in the image frame of video, it's hard to say some way transparency obtained
Mask is unique correct.Theoretically, above-described embodiment fusion more information, consideration are more multifactor, are conducive to more comprehensive
Image in video is investigated, to extract relatively satisfied video foreground target.It is understood that in above-described embodiment
In, it, can also be with when being extracted according to the foreground target in the first image of the first transparency mask to the video
It uses for reference, comprehensive related means in the prior art.That is, the key of above-described embodiment is to obtain how in new ways
Transparency mask and the trajectory extraction for implementing final specified target, and how do not lie according to transparency mask extraction video foreground
Target.
In another embodiment, the specified target derives from a photo, or derives from an image data base.Example
Such as, specified target is suspect, and photo is the recent photograph of the suspect, and image data base is wanted circular personnel image data
Library.
In another embodiment, further include following steps after the step S900:
S1000 extracts remaining each frame image from the video, and respectively as the first image, weight
Abovementioned steps S100 to S900 is executed, again to extract all foreground targets of the video；Or
S1100 extracts remaining each frame image from the video, respectively as the first image, and root
It is divided according to the first transparency mask of revised first image of previous frame all in the first image corresponding to the present frame
Foreground pixel set Fc, all background pixel set BCWith all unknown pixel set ZC, repeat abovementioned steps S200 extremely
S900, to extract all foreground targets of the video, wherein before dividing all in the first image corresponding to the present frame
Scape pixel set Fc, all background pixel set BCWith all unknown pixel set ZCSpecifically includes the following steps:
First transparency mask of revised first image of previous frame is carried out binaryzation by S11001, and threshold value takes 0.5,
Obtain the first bianry image of foreground target；
S11002, using the first bianry image as the second bianry image initial value；
S11003 carries out morphological erosion operation to the second bianry image using the circular configuration element that size is 3x3, and
The second bianry image is updated with the result of acquisition:
S11004 is repeated step S1003 five times；
S11005, using the first bianry image as third bianry image initial value；
S11006 carries out morphological dilation to third bianry image using the circular configuration element that size is 3x3, and
Third bianry image is updated with the result of acquisition:
S11007 is repeated step S1006 five times；
S11008 will be genuine respective pixel in the second bianry image as all foreground pixel set Fc, by the three or two
It is false respective pixel in value image as all background pixel set BC, rest of pixels is as all unknown pixel set ZC。
It is understood that repeating above-mentioned steps S100 to S900 to each frame image in video, view will be extracted
All foreground targets in frequency.But, it is contemplated that often each frame image and its a later frame image have in picture video pictures
Continuity and similitude in appearance, therefore, in order to make full use of this continuity and similitude, above-described embodiment can also be with
The institute in the first image corresponding to present frame is divided according to the first transparency mask of revised first image of previous frame
There is foreground pixel set Fc, all background pixel set BCWith all unknown pixel set ZC, so as in the essence of image procossing
Balance is obtained between degree and efficiency；That is, the embodiment has the characteristic inherited: inherit the transparent of former frame
Mask is spent, and divides foreground pixel set, background pixel set and the unknown pixel collection of a later frame using the transparency mask
Close, in view of the continuity and similitude on image content, therefore this division not only in accordance with the transparency mask of former frame and
And the means of morphological erosion and morphological dilations are utilized, this belongs to an innovative point of the disclosure.
In another embodiment, in step S600, in the following way to the first image superposition grayscale information to generate
Second image:
S601 carries out mean filter to the first image and obtains third image；
S602, the first image and third image generate the second image by following formula:
Wherein, IM2Indicate the gray value of k-th of pixel on the second image after being superimposed, xrIndicate k-th of picture on the first image
Plain xkNeighborhood territory pixel, NkIt indicates with xkCentered on neighborhood in number of pixels,Indicate to the first image into
The pixel value of k-th of pixel, β take 0.5 on the resulting third image of row mean filter.
The mode of specific superposition grayscale information is given by empirical value and related formula for above-described embodiment.
In another embodiment, step S800 further include:
S801 is found respectively according to the first transparency mask of the second transparency mask of the first image and the first image
The edge at the edge of its second transparency mask, the first transparency mask；
S802 obtains the position of all pixels at the edge of the second transparency mask and the edge of the first transparency mask
All pixels position, and determine position and the first transparency mask of all pixels at the edge of the second transparency mask
The region that the position of all pixels at edge is overlapped, and then determine the identical pixel Z in positionsp；
S803 searches pixel Z respectivelyspThe transparency estimated value of the first transparency mask corresponding to the first image and right
It should be in the transparency estimated value of the second transparency mask of the first image, and using the average value of the two as pixel ZspIt is revised
Transparency estimated value；
S804, with pixel ZspRevised transparency estimated value corrects the first transparency mask of the first image.
For above-described embodiment, it is intended to find, compares the identical pixel in position in two kinds of transparency masks, and utilize
Transparency estimated value of the identical pixel in the position in respective transparency mask is averaged to correct the of the first image
One transparency mask.
In another embodiment, the step S802 further comprises:
S8021, according to the position of all pixels at the edge of the second transparency mask of judgement and the first transparency mask
Edge all pixels position be overlapped region, further determine that the different pixel Z in positiondp, including two kinds of situations: it is located at
The pixel Z at the edge of the second transparency maskdp2With the pixel Z at the edge for being located at the first transparency maskdp1；
Unlike previous embodiment, edge that two transparency masks of the present embodiment additional attention are determined
The different pixel in middle position, and find out these pixels of position different from each other；
S8022 utilizes the different pixel Z in the positiondpPixel Z identical with positionsp, obtain the second transparency mask
Edge and the first transparency mask edge determined by: the closed enclosed region of institute and described between edge and edge
The position of all closing pixels of enclosed region；
For the step, the edge as corresponding to each mask can be considered as a connection or closure to a certain degree
Curve, then no matter closed curve corresponding to two masks is what kind of overlapping or nonoverlapping relationship: two are hidden
Those of on the corresponding edge of cover for the pixel of position not corresponding (i.e. position is different or position is not overlapped), jointly really
All closing pixels of the closed enclosed region of institute and the enclosed region between the edge and edge of two masks are determined
Position；
S8023 executes following sub-step:
(1) pixel Z is searcheddp1Position corresponding to pixel estimate in the transparency of the first transparency mask of the first image
Evaluation, and the transparence value of the corresponding pixel in the second image is searched, and using the average value of the two as pixel Zdp1Amendment
Transparency estimated value afterwards；
(2) pixel Z is searcheddp2Position corresponding to pixel estimate in the transparency of the second transparency mask of the first image
Evaluation, and the transparence value of the corresponding pixel in the first image is searched, and using the average value of the two as pixel Zdp2Amendment
Transparency estimated value afterwards；
For the step, it is transparent under two different systems to be intended to find each pixel in aforementioned enclosed region
Estimated value or transparence value are spent, and using the average value of the two as the revised transparency estimated value of respective pixel, then under
For correcting the first transparency mask of the first image in one step S8024.That is, the present embodiment is similar to previous reality
The amendment thinking for applying example is such, and what only the present embodiment solved is the common closed region of the corresponding edge institute of two masks.
Wherein, with pixel Zdp1For, belong to the pixel of the first transparency mask of the first image, first in the first image is saturating
There are a transparency estimated values for lightness mask, in addition, in the second image, pixel Zdp1In second image corresponding to position
Pixel have a transparence value in the second image, the present embodiment using the average value of the transparency estimated value and transparence value as
Respective pixel Zdp1Revised transparency estimated value.Pixel Zdp1It is similar.
S8024, in conjunction with pixel Zdp1Revised transparency estimated value and pixel Zdp2Revised transparency estimated value, is repaired
First transparency mask of positive the first image.For example, by pixel Zdp1Revised transparency estimated value and pixel Zdp2It repairs
Transparency estimated value after just, the transparence value as the first transparency mask corresponding position pixel.
Step in embodiment of the disclosure method can be sequentially adjusted, merged and deleted according to actual needs.
It should be noted that for the various method embodiments described above, for simple description, therefore, it is stated as a series of
Combination of actions, but those skilled in the art should understand that, the present invention is not limited by the sequence of acts described because
According to the present invention, some steps may be performed in other sequences or simultaneously.
In addition, referring to fig. 2, the disclosure further discloses a kind of video object trajectory extraction device in another embodiment,
Include:
First division module, is used for: for the first image in video, dividing all foreground pixel set in the image
F, all background pixel set B and all unknown pixel set Z；Wherein, the first image is extracted from the video
A certain frame image；
First metric module, is used for: giving certain prospect background pixels to (Fi, Bj), measurement is each not according to the following formula
Know pixel ZkTransparency
Wherein, IkFor unknown pixel ZkRGB color value, the foreground pixel FiFor apart from unknown pixel ZkNearest m
Foreground pixel, the background pixel BjAlso for apart from unknown pixel ZkM nearest background pixel, the prospect background pixel pair
(Fi, Bj) amount to m2Group；
Second metric module, is used for: for the m2Each group of prospect background pixel in group is to (Fi, Bj) and its it is corresponding
'sMeasurement prospect background pixel is to (F according to the following formulai, Bj) confidence level nij:
Wherein, σ value 0.1, and choose the highest MAX (n of confidence levelij) corresponding to that group prospect background pixel to for
(FiMAX, BjMAX)；
Computing module is used for: calculating each unknown pixel Z according to the following formulakTransparency estimated value
Determining module is used for: according to each unknown pixel ZkTransparency estimated valuePrimarily determine described first
First transparency mask of image；
Second division module, is used for: to the first image superposition grayscale information to generate the second image, and to second figure
As dividing its all foreground pixel set, all background pixel set and all unknown pixel set；
Calling module again is used for: being directed to second image, is called first metric module, the second measurement again
Module, computing module and determining module, to determine the first transparency mask of the second image, and by the first of second image
Second transparency mask of the transparency mask as the first image；
Correction module is used for: using the second transparency mask of the first image, correcting the first of the first image
Transparency mask；
Extraction module is used for: according to the first transparency mask of resulting first image of correction module, to the video
Foreground target in first image extracts, and specified target is retrieved in all foreground targets, then according to all frames
Time sequencing generates the track of the specified target according to all images for including the specified target.
For the embodiment, as shown in Figure 2, above-mentioned modules can be constituted with processor and memory system with
Just implement；But Fig. 2 is not interfered: modules can also have processing unit with itself to realize data-handling capacity.
In another embodiment, described device further includes following module:
Successively calling module is used for: remaining each frame image is extracted from the video, and respectively as described
First image, successively described in calling: the first division module, the second metric module, computing module, determines mould at the first metric module
Block, the second division module, again calling module, correction module and extraction module, to extract all foreground targets of the video；
Or include:
Calling module is inherited, is used for: extracting remaining each frame image from the video, respectively as described the
One image, and it is input to third division module, wherein the third division module is used for according to revised first figure of previous frame
First transparency mask of picture divides all foreground pixel set F in the first image corresponding to the present framec, had powerful connections
Pixel set BCWith all unknown pixel set ZC；Then the succession calling module successively calls first metric module,
Two metric modules, computing module, determining module, the second division module, again calling module, correction module and extraction module, with
Extract all foreground targets of the video, wherein third division module includes:
First binary Images Processing unit, for carrying out the first transparency mask of revised first image of previous frame
Binaryzation, threshold value take 0.5, obtain the first bianry image of foreground target；
Second bianry image initial cell, is used for: using the first bianry image as the second bianry image initial value；
Second binary Images Processing unit, is used for: using size be 3x3 circular configuration element to the second bianry image into
The operation of row morphological erosion, and the second bianry image is updated with the result obtained；
First repeats call unit, calls the second two-value processing unit five times for repeating；
Third bianry image initial cell, is used for: using the first bianry image as third bianry image initial value；
Third binary Images Processing unit, is used for: using size be 3x3 circular configuration element to third bianry image into
Row morphological dilation, and third bianry image is updated with the result obtained:
First repeats call unit, calls third two-value processing unit five times for repeating；
True and false division unit, is used for: will be true in the second bianry image of the second binary Images Processing unit final updating
Respective pixel as all foreground pixel set Fc, by the third bianry image of third binary Images Processing unit final updating
In be false respective pixel as all background pixel set BC, rest of pixels is as all unknown pixel set ZC。
In another embodiment, wherein the second division module further include:
Mean filter unit, is used for: carrying out mean filter to the first image and obtains third image；
Second image generation unit, is used for: the first image and third image pass through following formula the second image of generation:
Wherein, IM2Indicate the gray value of k-th of pixel on the second image after being superimposed, xrIndicate k-th of picture on the first image
Plain xkNeighborhood territory pixel, NkIt indicates with xkCentered on neighborhood in number of pixels,Indicate to the first image into
The pixel value of k-th of pixel, β take 0.5 on the resulting third image of row mean filter.
In another embodiment, wherein correction module further include:
Edge cells are found, are used for: according to the second transparency mask of the first image and the first transparency of the first image
Mask finds the edge of its second transparency mask, the edge of the first transparency mask respectively；
It determines position units, is used for: obtaining the position of all pixels at the edge of the second transparency mask and first transparent
The position of all pixels at the edge of mask is spent, and determines the position and first of all pixels at the edge of the second transparency mask
The region that the position of all pixels at the edge of transparency mask is overlapped, and then determine the identical pixel Z in positionsp；
First amending unit, is used for: searching pixel Z respectivelyspThe first transparency mask corresponding to the first image it is transparent
Spend estimated value, and the transparency estimated value of the second transparency mask corresponding to the first image, and using the average value of the two as
Pixel ZspRevised transparency estimated value；
Second amending unit, is used for: with pixel ZspRevised transparency estimated value, corrects the first of the first image
Transparency mask.
It is understood that described device can implement method described in one embodiment above.
In another embodiment, wherein the determining position units further comprise:
Different location subelement, is used for: according to the position of all pixels at the edge of the second transparency mask of judgement and
The region that the position of all pixels at the edge of the first transparency mask is overlapped, further determines that the different pixel Z in positiondp, packet
It includes: the pixel Z positioned at the edge of the second transparency maskdp2With the pixel Z at the edge for being located at the first transparency maskdp1；
It is closed subelement, is used for: utilizing the different pixel Z in the positiondpPixel Z identical with positionsp, obtain second
Determined by the edge of the edge of transparency mask and the first transparency mask: the closed closed area of institute between edge and edge
The position of all closing pixels of domain and the enclosed region；
Subelement is repeatedly searched, is used for:
(3) pixel Z is searcheddp1Position corresponding to pixel estimate in the transparency of the first transparency mask of the first image
Evaluation, and the transparence value of the corresponding pixel in the second image is searched, and using the average value of the two as pixel Zdp1Amendment
Transparency estimated value afterwards；
(4) pixel Z is searcheddp2Position corresponding to pixel estimate in the transparency of the second transparency mask of the first image
Evaluation, and the transparence value of the corresponding pixel first transparency mask in the first image is searched, and with the average value of the two
As pixel Zdp2Revised transparency estimated value；
Complicated revise subelemen, is used for: in conjunction with pixel Zdp1Revised transparency estimated value and pixel Zdp2It is revised
Transparency estimated value corrects the first transparency mask of the first image.
Those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, it is involved
And movement, module, unit be not necessarily essential to the invention.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, there is no the portion being described in detail in some embodiment
Point, reference can be made to the related descriptions of other embodiments.
In several embodiments provided by the disclosure, it should be understood that disclosed method is, it can be achieved that be corresponding function
Energy unit, processor or even system may be distributed over multiple wherein each section of the system both can be located in one place
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.In addition, each functional unit can integrate in one processing unit, it is also possible to each unit individualism, it can also two
A or more than two units are integrated in one unit.Above-mentioned integrated unit both can take the form of hardware realization, can also
To realize in the form of software functional units.If the integrated unit is realized in the form of SFU software functional unit and conduct
Independent product when selling or using, can store in a computer readable storage medium.Based on this understanding, originally
The disclosed technical solution substantially all or part of the part that contributes to existing technology or the technical solution in other words
It can be embodied in the form of software products, which is stored in a storage medium, including several fingers
It enables and using so that a computer equipment (can be smart phone, personal digital assistant, wearable device, laptop, plate
Computer) execute the disclosure each embodiment the method all or part of the steps.And storage medium above-mentioned include: USB flash disk,
Read-only memory (ROM, Read-Only Memory), is moved random access memory (RAM, Random Access Memory)
The various media that can store program code such as dynamic hard disk, magnetic or disk.
The above, above embodiments are only to illustrate the technical solution of the disclosure, rather than its limitations；Although referring to before
Embodiment is stated the disclosure is described in detail, it should be understood by those skilled in the art that: it still can be to aforementioned each reality
Technical solution documented by example is applied to modify or equivalent replacement of some of the technical features；And these modification or
Person's replacement, the range for the presently disclosed embodiments technical solution that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of video object track extraction method, includes the following steps:
S100 divides all foreground pixel set F in the image, all background pixel collections for the first image in video
Close B and all unknown pixel set Z；Wherein, the first image is a certain frame image extracted from the video；
S200 gives certain prospect background pixels to (Fi, Bj), each unknown pixel Z is measured according to the following formulakTransparency
Wherein, IkFor unknown pixel ZkRGB color value, the foreground pixel FiFor apart from unknown pixel ZkM nearest prospect
Pixel, the background pixel BjAlso for apart from unknown pixel ZkM nearest background pixel, the prospect background pixel is to (Fi,
Bj) amount to m2Group；
S300, for the m2Each group of prospect background pixel in group is to (Fi, Bj) and its it is correspondingAccording to the following formula
Prospect background pixel is measured to (Fi, Bj) confidence level nij:
Wherein, σ value 0.1, and choose the highest MAX (n of confidence levelij) corresponding to that group prospect background pixel to for
(FiMAX, BjMAX)；；
S400 calculates each unknown pixel Z according to the following formulakTransparency estimated value
S500, according to each unknown pixel ZkTransparency estimated valuePrimarily determine the first transparent of the first image
Spend mask；
S600 to the first image superposition grayscale information to generate the second image, and divides its all prospect to second image
Pixel set, all background pixel set and all unknown pixel set；
S700 executes step S200 to S500 for second image, to determine the first transparency mask of the second image,
And using the first transparency mask of second image as the second transparency mask of the first image；
S800 corrects the first transparency mask of the first image using the second transparency mask of the first image；
S900 corrects the first transparency mask of resulting first image according to step S800, to the first image of the video
In foreground target extract, and specified target is retrieved in all foreground targets, then according to the time sequencing of all frames,
The track of the specified target is generated according to all images for including the specified target.
2. according to the method described in claim 1, wherein, it is preferred that further include following steps after the step S900:
S1000 extracts remaining each frame image from the video, and respectively as the first image, repetition is held
Row abovementioned steps S100 to S900, to extract all foreground targets of the video；Or
S1100 extracts remaining each frame image, respectively as the first image, and according to upper from the video
First transparency mask of revised first image of one frame divides all prospects in the first image corresponding to the present frame
Pixel set Fc, all background pixel set BCWith all unknown pixel set ZC, abovementioned steps S200 to S900 is repeated,
To extract all foreground targets of the video, wherein divide all prospect pictures in the first image corresponding to the present frame
Plain set Fc, all background pixel set BCWith all unknown pixel set ZCSpecifically includes the following steps:
First transparency mask of revised first image of previous frame is carried out binaryzation by S11001, and threshold value takes 0.5, obtains
First bianry image of foreground target；
S11002, using the first bianry image as the second bianry image initial value；
S11003 carries out morphological erosion operation to the second bianry image using the circular configuration element that size is 3x3, and with obtaining
The result obtained updates the second bianry image:
S11004 is repeated step S1003 five times；
S11005, using the first bianry image as third bianry image initial value；
S11006 carries out morphological dilation to third bianry image using the circular configuration element that size is 3x3, and with obtaining
The result obtained updates third bianry image:
S11007 is repeated step S1006 five times；
S11008 will be genuine respective pixel in the second bianry image as all foreground pixel set Fc, by third bianry image
In be false respective pixel as all background pixel set BC, rest of pixels is as all unknown pixel set ZC。
3. according to the method described in claim 1, wherein, in step S600, in the following way to the first image superposition gray scale
Information is to generate the second image:
S601 carries out mean filter to the first image and obtains third image；
S602, the first image and third image generate the second image by following formula:
Wherein, IM2Indicate the gray value of k-th of pixel on the second image after being superimposed, xrIndicate k-th of pixel x on the first imagek
Neighborhood territory pixel, NkIt indicates with xkCentered on neighborhood in number of pixels,It indicates to carry out the first image equal
The pixel value of k-th of pixel, β take 0.5 on the resulting third image of value filtering.
4. according to the method described in claim 1, wherein, step S800 further include:
S801, according to the first transparency mask of the second transparency mask of the first image and the first image, find respectively its
The edge at the edge of two transparency masks, the first transparency mask；
S802 obtains the institute of the position of all pixels at the edge of the second transparency mask and the edge of the first transparency mask
There is the position of pixel, and determines the position of all pixels at the edge of the second transparency mask and the edge of the first transparency mask
All pixels the region that is overlapped of position, and then determine the identical pixel Z in positionsp；
S803 searches pixel Z respectivelyspThe transparency estimated value of the first transparency mask corresponding to the first image, and correspond to
The transparency estimated value of second transparency mask of the first image, and using the average value of the two as pixel ZspIt is revised transparent
Spend estimated value；
S804, with pixel ZspRevised transparency estimated value corrects the first transparency mask of the first image.
5. according to the method described in claim 4, wherein, the step S802 further comprises:
S8021, according to the side of the position of all pixels at the edge of the second transparency mask of judgement and the first transparency mask
The region that the position of all pixels of edge is overlapped, further determines that the different pixel Z in positiondp, comprising: it is located at the second transparency and hides
The pixel Z at the edge of coverdp2With the pixel Z at the edge for being located at the first transparency maskdp1；
S8022 utilizes the different pixel Z in the positiondpPixel Z identical with positionsp, obtain the edge of the second transparency mask
Determined by edge with the first transparency mask: the closed enclosed region of institute and the closed area between edge and edge
The position of all closing pixels in domain；
S8023 executes following sub-step:
(1) pixel Z is searcheddp1Position corresponding to pixel in the transparency estimated value of the first transparency mask of the first image,
And the transparence value of the corresponding pixel in the second image is searched, and using the average value of the two as pixel Zdp1It is revised
Transparency estimated value；
(2) pixel Z is searcheddp2Position corresponding to pixel in the transparency estimated value of the second transparency mask of the first image,
And the transparence value of the corresponding pixel in the first image is searched, and using the average value of the two as pixel Zdp2It is revised
Transparency estimated value；
S8024, in conjunction with pixel Zdp1Revised transparency estimated value and pixel Zdp2Revised transparency estimated value corrects institute
State the first transparency mask of the first image.
6. a kind of video object trajectory extraction device, comprising:
First division module, is used for: for the first image in video, dividing all foreground pixel set F in the image, institute
The pixel set that has powerful connections B and all unknown pixel set Z；Wherein, the first image be extracted from the video it is a certain
Frame image；
First metric module, is used for: giving certain prospect background pixels to (Fi, Bj), each unknown picture is measured according to the following formula
Plain ZkTransparency
Wherein, IkFor unknown pixel ZkRGB color value, the foreground pixel FiFor apart from unknown pixel ZkM nearest prospect
Pixel, the background pixel BjAlso for apart from unknown pixel ZkM nearest background pixel, the prospect background pixel is to (Fi,
Bj) amount to m2Group；
Second metric module, is used for: for the m2Each group of prospect background pixel in group is to (Fi, Bj) and its it is correspondingMeasurement prospect background pixel is to (F according to the following formulai, Bj) confidence level nij:
Wherein, σ value 0.1, and choose the highest MAX (n of confidence levelij) corresponding to that group prospect background pixel to for
(FiMAX, BjMAX)；
Computing module is used for: calculating each unknown pixel Z according to the following formulakTransparency estimated value
Determining module is used for: according to each unknown pixel ZkTransparency estimated valuePrimarily determine the first image
The first transparency mask；
Second division module, is used for: to the first image superposition grayscale information to generate the second image, and drawing to second image
Divide its all foreground pixel set, all background pixel set and all unknown pixel set；
Calling module again is used for: be directed to second image, call again first metric module, the second metric module,
Computing module and determining module, to determine the first transparency mask of the second image, and it is transparent by the first of second image
Spend second transparency mask of the mask as the first image；
Correction module is used for: using the second transparency mask of the first image, correct the first image first is transparent
Spend mask；
Extraction module is used for: according to the first transparency mask of resulting first image of correction module, to the first of the video
Foreground target in image extracts, and specified target is retrieved in all foreground targets, then according to the time of all frames
Sequentially, the track of the specified target is generated according to all images for including the specified target.
7. device according to claim 6, described device further include:
Successively calling module is used for: remaining each frame image is extracted from the video, and respectively as described first
Image, successively call described in: the first division module, the first metric module, the second metric module, computing module, determining module, the
Two division modules, again calling module, correction module and extraction module, to extract all foreground targets of the video；Or
Include:
Calling module is inherited, is used for: remaining each frame image is extracted from the video, respectively as first figure
Picture, and it is input to third division module, wherein the third division module is used for according to revised first image of previous frame
First transparency mask divides all foreground pixel set F in the first image corresponding to the present framec, all background pixels
Set BCWith all unknown pixel set ZC；Then the succession calling module successively calls first metric module, second degree
Module, computing module, determining module, the second division module, again calling module, correction module and extraction module are measured, to extract
All foreground targets of the video, wherein third division module includes:
First binary Images Processing unit, for the first transparency mask of revised first image of previous frame to be carried out two-value
Change, threshold value takes 0.5, obtains the first bianry image of foreground target；
Second bianry image initial cell, is used for: using the first bianry image as the second bianry image initial value；
Second binary Images Processing unit, is used for: carrying out shape to the second bianry image using the circular configuration element that size is 3x3
State etching operation, and the second bianry image is updated with the result obtained；
First repeats call unit, calls the second two-value processing unit five times for repeating；
Third bianry image initial cell, is used for: using the first bianry image as third bianry image initial value；
Third binary Images Processing unit, is used for: carrying out shape to third bianry image using the circular configuration element that size is 3x3
State expansive working, and third bianry image is updated with the result obtained:
First repeats call unit, calls third two-value processing unit five times for repeating；
True and false division unit, is used for: will be really right in the second bianry image of the second binary Images Processing unit final updating
Answer pixel as all foreground pixel set Fc, will be in the third bianry image of third binary Images Processing unit final updating
False respective pixel is as all background pixel set BC, rest of pixels is as all unknown pixel set ZC。
8. device according to claim 6, wherein the second division module further include: mean filter unit is used for: to
One image carries out mean filter and obtains third image；
Second image generation unit, is used for: the first image and third image pass through following formula the second image of generation:
Wherein, IM2Indicate the gray value of k-th of pixel on the second image after being superimposed, xrIndicate k-th of pixel x on the first imagek
Neighborhood territory pixel, NkIt indicates with xkCentered on neighborhood in number of pixels,It indicates to carry out the first image equal
The pixel value of k-th of pixel, β take 0.5 on the resulting third image of value filtering.
9. device according to claim 6, wherein correction module further include:
Edge cells are found, are used for: according to the first transparency mask of the second transparency mask of the first image and the first image,
The edge of its second transparency mask, the edge of the first transparency mask are found respectively；
Determine position units, be used for: the position and the first transparency for obtaining all pixels at the edge of the second transparency mask hide
The position of all pixels at the edge of cover, and determine the position of all pixels at the edge of the second transparency mask and first transparent
The region that the position of all pixels at the edge of mask is overlapped is spent, and then determines the identical pixel Z in positionsp；
First amending unit, is used for: searching pixel Z respectivelyspThe transparency of the first transparency mask corresponding to the first image is estimated
Evaluation, and the transparency estimated value of the second transparency mask corresponding to the first image, and using the average value of the two as pixel
ZspRevised transparency estimated value；
Second amending unit, is used for: with pixel ZspRevised transparency estimated value, correct the first image first are transparent
Spend mask.
10. device according to claim 9, wherein the determining position units further comprise:
Different location subelement, is used for: according to the position and first of all pixels at the edge of the second transparency mask of judgement
The region that the position of all pixels at the edge of transparency mask is overlapped, further determines that the different pixel Z in positiondp, comprising: position
Pixel Z in the edge of the second transparency maskdp2With the pixel Z at the edge for being located at the first transparency maskdp1；
It is closed subelement, is used for: utilizing the different pixel Z in the positiondpPixel Z identical with positionsp, obtain the second transparency
Determined by the edge of the edge of mask and the first transparency mask: the closed enclosed region of institute between edge and edge, and
The position of all closing pixels of the enclosed region；
Subelement is repeatedly searched, is used for:
(1) pixel Z is searcheddp1Position corresponding to pixel in the transparency estimated value of the first transparency mask of the first image,
And the transparence value of the corresponding pixel in the second image is searched, and using the average value of the two as pixel Zdp1It is revised
Transparency estimated value；
(2) pixel Z is searcheddp2Position corresponding to pixel in the transparency estimated value of the second transparency mask of the first image,
And the transparence value of the corresponding pixel in the first image is searched, and using the average value of the two as pixel Zdp2It is revised
Transparency estimated value；
Complicated revise subelemen, is used for: in conjunction with pixel Zdp1Revised transparency estimated value and pixel Zdp2It is revised transparent
Estimated value is spent, the first transparency mask of the first image is corrected.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNPCT/CN2018/107514 | 2018-09-26 | ||
| CN2018107514 | 2018-09-26 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110363788A true CN110363788A (en) | 2019-10-22 |
Family
ID=68140393
Family Applications (5)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910444633.4A Withdrawn CN110378867A (en) | 2018-09-26 | 2019-05-24 | By prospect background pixel to and grayscale information obtain transparency mask method |
| CN201910444632.XA Withdrawn CN110335288A (en) | 2018-09-26 | 2019-05-24 | A kind of video foreground target extraction method and device |
| CN201910505008.6A Withdrawn CN110363788A (en) | 2018-09-26 | 2019-06-11 | A kind of video object track extraction method and device |
| CN201910628287.5A Withdrawn CN110516534A (en) | 2018-09-26 | 2019-07-11 | A kind of method for processing video frequency and device based on semantic analysis |
| CN201910737589.6A Withdrawn CN110659562A (en) | 2018-09-26 | 2019-08-09 | Deep learning (DNN) classroom learning behavior analysis method and device |
Family Applications Before (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910444633.4A Withdrawn CN110378867A (en) | 2018-09-26 | 2019-05-24 | By prospect background pixel to and grayscale information obtain transparency mask method |
| CN201910444632.XA Withdrawn CN110335288A (en) | 2018-09-26 | 2019-05-24 | A kind of video foreground target extraction method and device |
Family Applications After (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910628287.5A Withdrawn CN110516534A (en) | 2018-09-26 | 2019-07-11 | A kind of method for processing video frequency and device based on semantic analysis |
| CN201910737589.6A Withdrawn CN110659562A (en) | 2018-09-26 | 2019-08-09 | Deep learning (DNN) classroom learning behavior analysis method and device |
Country Status (2)
| Country | Link |
|---|---|
| CN (5) | CN110378867A (en) |
| WO (5) | WO2020062899A1 (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989962B (en) * | 2021-02-24 | 2024-01-05 | 上海商汤智能科技有限公司 | Track generation method, track generation device, electronic equipment and storage medium |
| KR20240005727A (en) * | 2021-04-06 | 2024-01-12 | 나이앤틱, 인크. | Panoptic segmentation prediction for augmented reality |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102456212A (en) * | 2010-10-19 | 2012-05-16 | 北大方正集团有限公司 | Separation method and system for visible watermark in numerical image |
| CN102999892A (en) * | 2012-12-03 | 2013-03-27 | 东华大学 | Intelligent fusion method for depth images based on area shades and red green blue (RGB) images |
| CN103366364A (en) * | 2013-06-07 | 2013-10-23 | 太仓中科信息技术研究院 | Color difference-based image matting method |
| US8731315B2 (en) * | 2011-09-12 | 2014-05-20 | Canon Kabushiki Kaisha | Image compression and decompression for image matting |
| WO2015048694A2 (en) * | 2013-09-27 | 2015-04-02 | Pelican Imaging Corporation | Systems and methods for depth-assisted perspective distortion correction |
| US20170116481A1 (en) * | 2015-10-23 | 2017-04-27 | Beihang University | Method for video matting via sparse and low-rank representation |
| CN107516319A (en) * | 2017-09-05 | 2017-12-26 | 中北大学 | A kind of high accuracy simple interactive stingy drawing method, storage device and terminal |
| CN108391118A (en) * | 2018-03-21 | 2018-08-10 | 惠州学院 | A kind of display system for realizing 3D rendering based on projection pattern |
Family Cites Families (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6361910B1 (en) * | 2000-02-03 | 2002-03-26 | Applied Materials, Inc | Straight line defect detection |
| US6870945B2 (en) * | 2001-06-04 | 2005-03-22 | University Of Washington | Video object tracking by estimating and subtracting background |
| US7466842B2 (en) * | 2005-05-20 | 2008-12-16 | Mitsubishi Electric Research Laboratories, Inc. | Modeling low frame rate videos with bayesian estimation |
| US8508546B2 (en) * | 2006-09-19 | 2013-08-13 | Adobe Systems Incorporated | Image mask generation |
| US8520972B2 (en) * | 2008-09-12 | 2013-08-27 | Adobe Systems Incorporated | Image decomposition |
| CN101686338B (en) * | 2008-09-26 | 2013-12-25 | 索尼株式会社 | System and method for partitioning foreground and background in video |
| CN101588459B (en) * | 2009-06-26 | 2011-01-05 | 北京交通大学 | Video keying processing method |
| CN101621615A (en) * | 2009-07-24 | 2010-01-06 | 南京邮电大学 | Self-adaptive background modeling and moving target detecting method |
| US8625888B2 (en) * | 2010-07-21 | 2014-01-07 | Microsoft Corporation | Variable kernel size image matting |
| US8386964B2 (en) * | 2010-07-21 | 2013-02-26 | Microsoft Corporation | Interactive image matting |
| CN102163216B (en) * | 2010-11-24 | 2013-02-13 | 广州市动景计算机科技有限公司 | Picture display method and device thereof |
| CN102236901B (en) * | 2011-06-30 | 2013-06-05 | 南京大学 | Method for tracking target based on graph theory cluster and color invariant space |
| US8744123B2 (en) * | 2011-08-29 | 2014-06-03 | International Business Machines Corporation | Modeling of temporarily static objects in surveillance video data |
| US9305357B2 (en) * | 2011-11-07 | 2016-04-05 | General Electric Company | Automatic surveillance video matting using a shape prior |
| CN102651135B (en) * | 2012-04-10 | 2015-06-17 | 电子科技大学 | Optimized direction sampling-based natural image matting method |
| US8792718B2 (en) * | 2012-06-29 | 2014-07-29 | Adobe Systems Incorporated | Temporal matte filter for video matting |
| AU2013206597A1 (en) * | 2013-06-28 | 2015-01-22 | Canon Kabushiki Kaisha | Depth constrained superpixel-based depth map refinement |
| US20150091891A1 (en) * | 2013-09-30 | 2015-04-02 | Dumedia, Inc. | System and method for non-holographic teleportation |
| CN104112144A (en) * | 2013-12-17 | 2014-10-22 | 深圳市华尊科技有限公司 | Person and vehicle identification method and device |
| US10089740B2 (en) * | 2014-03-07 | 2018-10-02 | Fotonation Limited | System and methods for depth regularization and semiautomatic interactive matting using RGB-D images |
| CN104952089B (en) * | 2014-03-26 | 2019-02-15 | 腾讯科技（深圳）有限公司 | A kind of image processing method and system |
| CN103903230A (en) * | 2014-03-28 | 2014-07-02 | 哈尔滨工程大学 | Video image sea fog removal and clearing method |
| CN105590307A (en) * | 2014-10-22 | 2016-05-18 | 华为技术有限公司 | Transparency-based matting method and apparatus |
| CN104573688B (en) * | 2015-01-19 | 2017-08-25 | 电子科技大学 | Mobile platform tobacco laser code intelligent identification Method and device based on deep learning |
| CN104680482A (en) * | 2015-03-09 | 2015-06-03 | 华为技术有限公司 | Method and device for image processing |
| CN104935832B (en) * | 2015-03-31 | 2019-07-12 | 浙江工商大学 | For the video keying method with depth information |
| CN105100646B (en) * | 2015-08-31 | 2018-09-11 | 北京奇艺世纪科技有限公司 | Method for processing video frequency and device |
| CN105809679B (en) * | 2016-03-04 | 2019-06-18 | 李云栋 | Mountain railway side slope rockfall detection method based on visual analysis |
| US10275892B2 (en) * | 2016-06-09 | 2019-04-30 | Google Llc | Multi-view scene segmentation and propagation |
| CN106204567B (en) * | 2016-07-05 | 2019-01-29 | 华南理工大学 | A kind of natural background video matting method |
| CN107665326B (en) * | 2016-07-29 | 2024-02-09 | 奥的斯电梯公司 | Monitoring system for passenger conveyor, passenger conveyor and monitoring method thereof |
| CN107872644B (en) * | 2016-09-23 | 2020-10-09 | 亿阳信通股份有限公司 | Video monitoring method and device |
| CN106778810A (en) * | 2016-11-23 | 2017-05-31 | 北京联合大学 | Original image layer fusion method and system based on RGB feature Yu depth characteristic |
| US10198621B2 (en) * | 2016-11-28 | 2019-02-05 | Sony Corporation | Image-Processing device and method for foreground mask correction for object segmentation |
| CN106952276A (en) * | 2017-03-20 | 2017-07-14 | 成都通甲优博科技有限责任公司 | A kind of image matting method and device |
| CN107194867A (en) * | 2017-05-14 | 2017-09-22 | 北京工业大学 | A kind of stingy picture synthetic method based on CUDA |
| CN107273905B (en) * | 2017-06-14 | 2020-05-08 | 电子科技大学 | Target active contour tracking method combined with motion information |
| CN107230182B (en) * | 2017-08-03 | 2021-11-09 | 腾讯科技（深圳）有限公司 | Image processing method and device and storage medium |
| CN108399361A (en) * | 2018-01-23 | 2018-08-14 | 南京邮电大学 | A kind of pedestrian detection method based on convolutional neural networks CNN and semantic segmentation |
| CN108320298B (en) * | 2018-04-28 | 2022-01-28 | 亮风台（北京）信息科技有限公司 | Visual target tracking method and equipment |
-
2019
- 2019-05-24 WO PCT/CN2019/088279 patent/WO2020062899A1/en active Application Filing
- 2019-05-24 WO PCT/CN2019/088278 patent/WO2020062898A1/en active Application Filing
- 2019-05-24 CN CN201910444633.4A patent/CN110378867A/en not_active Withdrawn
- 2019-05-24 CN CN201910444632.XA patent/CN110335288A/en not_active Withdrawn
- 2019-06-11 CN CN201910505008.6A patent/CN110363788A/en not_active Withdrawn
- 2019-07-11 CN CN201910628287.5A patent/CN110516534A/en not_active Withdrawn
- 2019-08-09 CN CN201910737589.6A patent/CN110659562A/en not_active Withdrawn
- 2019-08-19 WO PCT/CN2019/101273 patent/WO2020063189A1/en active Application Filing
- 2019-09-10 WO PCT/CN2019/105028 patent/WO2020063321A1/en active Application Filing
- 2019-09-19 WO PCT/CN2019/106616 patent/WO2020063436A1/en active Application Filing
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102456212A (en) * | 2010-10-19 | 2012-05-16 | 北大方正集团有限公司 | Separation method and system for visible watermark in numerical image |
| US8731315B2 (en) * | 2011-09-12 | 2014-05-20 | Canon Kabushiki Kaisha | Image compression and decompression for image matting |
| CN102999892A (en) * | 2012-12-03 | 2013-03-27 | 东华大学 | Intelligent fusion method for depth images based on area shades and red green blue (RGB) images |
| CN103366364A (en) * | 2013-06-07 | 2013-10-23 | 太仓中科信息技术研究院 | Color difference-based image matting method |
| WO2015048694A2 (en) * | 2013-09-27 | 2015-04-02 | Pelican Imaging Corporation | Systems and methods for depth-assisted perspective distortion correction |
| US20170116481A1 (en) * | 2015-10-23 | 2017-04-27 | Beihang University | Method for video matting via sparse and low-rank representation |
| CN107516319A (en) * | 2017-09-05 | 2017-12-26 | 中北大学 | A kind of high accuracy simple interactive stingy drawing method, storage device and terminal |
| CN108391118A (en) * | 2018-03-21 | 2018-08-10 | 惠州学院 | A kind of display system for realizing 3D rendering based on projection pattern |
Non-Patent Citations (4)
| Title |
|---|
| NING XU 等: "Deep Image Matting", 《COMPUTER VISION FOUNDATION》 * |
| SHUTAO LI 等: "Image matting for fusion of multi-focus images in dynamic scenes", 《INFORMATION FUSION》 * |
| ZHAOQUAN CAI 等: "Improving sampling-based image matting with cooperative coevolution differential evolution algorithm", 《SOFT COMPUTING》 * |
| 谢斌 等: "基于雾气遮罩理论的图像去雾算法", 《计算机工程与科学》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110335288A (en) | 2019-10-15 |
| CN110659562A (en) | 2020-01-07 |
| CN110516534A (en) | 2019-11-29 |
| WO2020062899A1 (en) | 2020-04-02 |
| WO2020063189A1 (en) | 2020-04-02 |
| WO2020062898A1 (en) | 2020-04-02 |
| WO2020063321A1 (en) | 2020-04-02 |
| WO2020063436A1 (en) | 2020-04-02 |
| CN110378867A (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3540637B1 (en) | Neural network model training method, device and storage medium for image processing | |
| CN111461110B (en) | Small target detection method based on multi-scale image and weighted fusion loss | |
| CN110047069B (en) | Image detection device | |
| CN108986140B (en) | Target scale self-adaptive tracking method based on correlation filtering and color detection | |
| CN110765860B (en) | Tumble judging method, tumble judging device, computer equipment and storage medium | |
| CN107169463B (en) | Method for detecting human face, device, computer equipment and storage medium | |
| CN111476806B (en) | Image processing method, image processing device, computer equipment and storage medium | |
| CN108961158B (en) | Image synthesis method and device | |
| CN112330684B (en) | Object segmentation method and device, computer equipment and storage medium | |
| CN113378812A (en) | Digital dial plate identification method based on Mask R-CNN and CRNN | |
| CN111783779A (en) | Image processing method, apparatus and computer-readable storage medium | |
| JP2007128195A (en) | Image processing system | |
| CN113870157A (en) | SAR image synthesis method based on cycleGAN | |
| CN116645592B (en) | Crack detection method based on image processing and storage medium | |
| CN114021704B (en) | AI neural network model training method and related device | |
| CN110390259A (en) | Recognition methods, device, computer equipment and the storage medium of diagram data | |
| CN116030498A (en) | Virtual garment running and showing oriented three-dimensional human body posture estimation method | |
| CN110363788A (en) | A kind of video object track extraction method and device | |
| CN111563462A (en) | Image element detection method and device | |
| CN114004772B (en) | Image processing method, and method, system and equipment for determining image synthesis model | |
| CN111027472A (en) | Video identification method based on fusion of video optical flow and image space feature weight | |
| CN117934827A (en) | Key point detection model training method, key point detection method and related device | |
| CN113506226B (en) | Motion blur restoration method and system | |
| CN110766079B (en) | Training data generation method and device for screen abnormal picture detection | |
| CN114648800A (en) | Face image detection model training method, face image detection method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20191022 |
|
| WW01 | Invention patent application withdrawn after publication |