DE102007033316A1 - Method for vectorizing video images, involves determining video image in sequential frame contours of image structures on basis of volatile changes of brightness or chromaticity in base analysis by neighboring pixels - Google Patents

Abstract

The method involves determining the video image in sequential frame contours of image structures on the basis of volatile changes of brightness or chromaticity in a base analysis by neighboring pixels. The contour of the object is described as a vector sequence and object with priorities are provided. Textures are stored to the object and textures with higher accuracy are determined according to the priority of the objects.

Description

The The invention relates to a method for vectoring video images, as a result of frames with two-dimensionally addressable pixels and associated pixel data, and after storage and / or transfer are largely regenerated again, in a basic analysis of the video images in successive Frames Contours of image structures based on sudden changes brightness and / or color value with neighboring pixels determined, defined from the contours so found objects become, for the individual objects a motion vector and color vectors can be determined from which objects a background is determined

Under the ISO standards MPEG1 to MPEG4, are methods of compression become known from video images. These go to a screening of the Image in 8x8 or 16x16 pixel blocks whose content is described and the residual error about the Discrete cosine transformation is specified. Such blocks can be grouped into objects whose movement in the picture with vectors are described. This results in only one coarse capture of image content to objects and the block structure remains visible on complex objects.

In the American patent US 6,026,182 For example, a method for dividing video images into objects for the purpose of compression is described by finding inner and outer perimeters of the objects. The movement of the objects to each other is estimated by means of anchor points (feature points). No information is given on the treatment of textures of the objects.

In the German patent DE 102 31 613 B4 For example, a method of compressing and decompressing video image data is described, which involves detecting objects within the video images and then processing those objects for compression and transmission.

It It is the object of the invention to continue the aforementioned method in the same way improve that further compression of the image data is enabled and that the playback of the video images of the original size of the images is independent and a quality improvement of the video images regarding the sharpness as well as the textures is made.

These The object is achieved by the characterizing features of claim 1 solved.

advantageous Embodiments of the method are in the subclaims specified.

The Basis for the image processing according to the invention is the object recognition, which is about the recognition of contours in successive frames. Typically, four Evaluated frames and to the detected objects, a motion vector and color vectors determined. The non-moving contours become the Assigned background. The contours recognized as objects become described as a vector sequence and with an identity number Mistake. Priorities are formed for the objects and saved to the object.

advantageously, only contours are taken over as objects in the image processing, which include a minimum number of pixels in the video image. This minimum number can be a fixed value, about 64 pixels. It can also be one relative value relative to the total number of pixels in the video image, about 0.1 per thousand. The shape of the contour does not matter at. For example, hair in the picture is also captured, as it lasts a long time Have extension and thus exceed the limit for the number of pixels. However, limiting the number of pixels will cause artifacts and image noise not further processed in the original image and thus effective suppressed.

It is convenient to assign the objects to image layers. A Image plane is the background on which the non-moving contours being represented.

The Priorities of the objects are advantageously dependent from the movement, the position in the picture, the picture plane, the part on the overall picture content and the sharpness. The bigger, sharper or faster an object is, the more important it is for the bigger picture and the higher it is his priority. The priority decides over the computational effort expended on an object and therefore about the accuracy and the quality.

The Surface of the objects is first with a Color dominance vector and provided with gradient vectors. By this vectorization is also effectively suppressing image noise.

at Patterns and irregular content will be this described as textures. The textures come with an index number provided and stored. Thereby the textures become according to Priority determined with higher accuracy.

complicated Textures are only roughly resolved and then refined with information from subsequent video images. Consequently the quality of rendering of textures increases.

Objects with the same texture grab over the Index number to the stored texture information. Thus, this information only needs to be transmitted once.

The Information about the objects, textures and background as a stream that stores or transmits can be. When playing back the video images will be on this data stream accessed and built the information about pixels in frames. Since the information is in vector form, frame size matters not play. The vectors are in the desired Size scales. That way you can Video images that are in a small frame size, converted to higher resolutions and larger be projected.

artifacts and image noise in the original images are suppressed and thus a higher quality image is displayed.

In the 1 the method of image processing is shown schematically.

The original video images 100 lie as a result of frames 150 with associated pixel data. From these frames 150 become in the object recognition 200 the contours are determined and assigned to the objects. At vectorization 300 the vectors are determined to the objects. These include the contours, colors and gradients. The textures are in the texture formation 400 described and further from the following frames of the video images 100 refined. The recovered texture information 450 will be in the memory 500 stored and can be taken to further text descriptions again.

From the information on the video images 100 becomes a vectorized stream 550 formed and for playback the decoder 800 fed.

Alternatively, the vectorized data stream 550 also the compressor 600 are fed, the compressed data stream 650 emits. This becomes the decompressor 700 supplied, decompressed and according to the transfer stream 550 in the decoder 800 decoded.

In the playback memory 900 become the playback frames 950 generated and as a regenerated video image 1000 output. This corresponds to the original video image 100 with improved quality, since artifacts and video noise are suppressed.

100

original video images

150

video frames

200

object recognition

300

vectorization

400

texturing

450

texture data

500

Storage

550

vectorized data stream

600

compressor

650

compressed data stream

700

decompressor

800

decoder

900

Playback Memory

950

Playback frames

1000

regenerated video images

QUOTES INCLUDE IN THE DESCRIPTION

This list The documents listed by the applicant have been automated generated and is solely for better information recorded by the reader. The list is not part of the German Patent or utility model application. The DPMA takes over no liability for any errors or omissions.

Cited patent literature

• - US 6026182 [0003]
• - DE 10231613 B4 [0004]

Claims (14)

A method for vectorizing video images which are present as a consequence of frames having two-dimensionally addressable pixels and associated pixel data, and which are largely regenerated after storage and / or transmission, wherein - in a basic analysis of the video images in successive frames, contours of image structures are determined by means of stepwise Changes in brightness and / or color value are determined in adjacent pixels, - objects are defined from the contours found, - a motion vector and color vectors are determined for the individual objects, - a background is determined from the objects, characterized in that - Contours of the objects are described as vector sequences, - the objects are prioritized, - textures are stored to the objects, and - the textures are determined according to the priority of the objects with higher accuracy. Method according to claim 1, characterized in that that only contours that include a minimum number of pixels than Objects are treated. Method according to claim 2, characterized in that that the minimum number is 64 pixels. Method according to claim 1, characterized in that that the objects are provided with an identity number. Method according to claim 1, characterized in that that the objects are assigned to image planes. Method according to claim 5, characterized in that that non-moving objects are assigned to the background. become. Method according to claim 1, characterized in that that the priorities of the objects - of the Move, - the position in the picture, - of the Image plane, - the proportion of the overall picture content, - of the sharpness be formed. Method according to claim 1, characterized in that that the quality of playing an object with his Priority is increasing. Method according to claim 1, characterized in that to the color vectors, a color dominance vector and color gradient vectors belong. Method according to claim 1, characterized in that that the textures of the objects initially coarser be resolved and refined from subsequent video images. Method according to claim 1, characterized in that that with known textures when playing the objects on the stored information is used. Method according to claim 1, characterized in that that transmitted the object data for as a data stream become. Method according to claim 12, characterized in that that regenerates the objects from the data stream and as video images be displayed in frames. Method according to claim 13, characterized in that that the frame size when playing back from the original frame size different is.