CN113763378B - Scoring method for license plate image non-reference quality analysis - Google Patents
Scoring method for license plate image non-reference quality analysis Download PDFInfo
- Publication number
- CN113763378B CN113763378B CN202111107282.1A CN202111107282A CN113763378B CN 113763378 B CN113763378 B CN 113763378B CN 202111107282 A CN202111107282 A CN 202111107282A CN 113763378 B CN113763378 B CN 113763378B
- Authority
- CN
- China
- Prior art keywords
- license plate
- plate image
- sample
- image
- image sample
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- 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
- G06T2207/30168—Image quality inspection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention provides a scoring method for license plate image non-reference quality analysis, which comprises the following steps: obtaining a license plate image sample total set; obtaining a license plate image sample total set; determining a threshold value of an effective scoring algorithm; and scoring one license plate image and outputting the total score of the image definition. The advantages are that: the scoring method solves the limitation that the full-reference image quality evaluation depends on a label image and the singleness of the non-reference image quality evaluation method, comprehensively considers the license plate image quality of various factors such as the image horizontal gradient, edge detection, gray variance, energy gradient, edge sharpness, secondary blurring, low-pass filtering, image structure similarity and the like, has more elements, and is more objective and accurate in evaluation.
Description
Technical Field
The invention belongs to the technical field of computer vision, and particularly relates to a scoring method for license plate image non-reference quality analysis.
Background
The existing image quality analysis techniques are mainly divided into no-reference image quality analysis and full-reference image quality analysis techniques. In the full-reference image quality analysis, the test image needs to adopt clear images (as labels) with the same image content as a group of paired images, so that the full-reference evaluation index of the test image can be calculated. The analysis method is limited by the establishment of a data set, and the image in the actual scene has no label image and cannot calculate the evaluation index of the full reference image.
According to the reference-free image quality analysis method, the image quality score can be calculated only according to the single image data of the actual scene. The method has the advantages that: there is no need to establish paired data sets and is not subject to image data that is not paired in a real-world scenario. However, the current reference-free image quality analysis method focuses on a single factor, such as the gradient of the brenner in the horizontal direction of the image, and cannot integrate factors in multiple aspects to evaluate the quality of the image, so that the accuracy of the quality score of the finally obtained image is limited.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a scoring method for license plate image non-reference quality analysis, which can effectively solve the problems.
The technical scheme adopted by the invention is as follows:
the invention provides a scoring method for license plate image non-reference quality analysis, which comprises the following steps:
step 1, obtaining a license plate image sample total set S;
step 1.1, obtaining m license plate images; preprocessing each license plate image, and processing each license plate image into a license plate image with uniform size as a license plate image sample;
obtaining m license plate image samples; each license plate image sample is provided with a sample label; the sample label comprises a clear license plate image label or a fuzzy license plate image label;
in the m Zhang Chepai image samples, m is included 1 Clear license plate image sample and m 2 A blurred license plate image sample is stretched;
step 1.2, m 1 The clear license plate image samples form a clear license plate image sample set S 1 The method comprises the steps of carrying out a first treatment on the surface of the Let m 2 The blurred license plate image samples form blurred license plate image sample set S 2 ；
Clear license plate image sample set S 1 And blurred license plate image sample set S 2 Forming a license plate image sample total set S;
step 2, constructing a license plate image effective scoring algorithm set, wherein the license plate image effective scoring algorithm set comprises n license plate image effective scoring algorithms, which are respectively expressed as: effective scoring algorithm SF 1 ,SF 2 ,...,SF n ；
The specific method comprises the following steps:
for each license plate image scoring algorithm G, identifying whether the license plate image scoring algorithm G is a license plate image effective scoring algorithm, if so, adding the license plate image scoring algorithm G into the license plate image effective scoring algorithm set;
wherein: the method is characterized in that whether the license plate image scoring algorithm G is a license plate image effective scoring algorithm is identified by adopting the following method:
step 2.1, respectively calculating clear license plate image sample sets S by adopting a license plate image scoring algorithm G 1 Image definition score of each clear license plate image sample in the license plate image is calculated to obtain m 1 Scoring the definition of each image;
step 2.2, for m 1 Counting the distribution of the individual image definition scores to obtain an image definition distribution map of a clear license plate image sample, wherein the abscissa is the image definition score and the ordinate is the sample number;
step 2.3, respectively calculating a fuzzy license plate image sample set S by adopting a license plate image scoring algorithm G 2 Image definition score of each blurred license plate image sample, thereby calculating m 2 Scoring the definition of each image;
step 2.4, for m 2 Counting the distribution of the individual image definition scores to obtain an image definition distribution map of a fuzzy license plate image sample with the abscissa being the image definition score and the ordinate being the sample number;
step 2.5, judging whether the image definition distribution map of the clear license plate image sample and the image definition distribution map of the fuzzy license plate image sample have obvious differences, if so, obtaining a conclusion that a license plate image scoring algorithm G is a license plate image effective scoring algorithm; otherwise, a conclusion that the license plate image scoring algorithm G is a license plate image invalid scoring algorithm is obtained;
step 3, for each effective scoring algorithm SF i Wherein i=1, 2, n, determining a corresponding algorithm threshold K i ：
Step 3.1, determining an effective scoring algorithm SF i Initial threshold K of (2) i (0):
SF is adopted by an effective scoring algorithm i Respectively calculating fuzzy license plate image sample sets S 2 The image definition score of each blurred license plate image sample is selected, and the image definition score with the lowest image definition score is used as an effective scoring algorithm SF i Initial threshold K of (2) i (0)；
Step 3.2, presetting a step rho;
step 3.3, from the initial threshold value K i (0) Starting, traversing by taking the step rho as an increment value to obtain a threshold meeting the following objective function as an effective scoring algorithm SF i Is the optimal algorithm threshold K of (2) i ：
Objective function:
the classification error rate does not exceed the set value epsilon; and, from the blurred license plate image sample set S 2 Identifying the number E of the fuzzy license plate images to be the largest;
wherein:
the classification error rate is calculated by the following method:
1) For each license plate image sample in the license plate image sample total set S, representing the license plate image sample as a license plate image sample P;
2) SF is adopted by an effective scoring algorithm i Identifying a license plate image sample P to obtain a classification result; wherein, the classification result is: the license plate image sample P is a blurred license plate image sample or a clear license plate image sample;
specifically, an effective scoring algorithm SF is adopted i Calculating to obtain an image definition Score (P) of the license plate image sample P;
comparing whether the image sharpness Score (P) is greater than the threshold value K of the current traversal i (0) +xρ, where x is the number of steps traversed;
if the number is larger than the number, obtaining a classification result that the license plate image sample P is a clear license plate image sample; otherwise, obtaining a classification result that the license plate image sample P is a fuzzy license plate image sample;
3) Comparing the classification result of the license plate image sample P with a sample label of the license plate image sample P, and if the classification result is consistent with the sample label, indicating that the classification of the license plate image sample P is correct; otherwise, indicating that the classification of the license plate image sample P is wrong;
4) Therefore, for each license plate image sample in the license plate image sample total set S, a classification result is obtained, and the classification error rate is calculated by adopting the following formula:
classification error rate = number of samples with classification errors/total number of license plate image samples of the license plate image sample total set S;
from a blurred license plate image sample set S 2 In the method, the number E of the fuzzy license plate images is identified, and the method is adopted for calculating the number E of the fuzzy license plate images:
1) For a blurred license plate image sample set S 2 Each fuzzy license plate image sample in the system adopts an effective scoring algorithm SF i Identifying the obtained product to obtain a classification result; wherein, the classification result is: the blurred license plate image sample is a blurred license plate image sample or a clear license plate image sample;
2) For a blurred license plate image sample set S 2 Comprising m 2 The fuzzy license plate image samples are expanded, and the number of the fuzzy license plate image samples is counted to obtain a classification result, namely the number E of the fuzzy license plate images is identified;
step 4, thereby determining the effective scoring algorithm SF 1 ,SF 2 ,...,SF n The algorithm thresholds of (a) are respectively: k (K) 1 ,K 2 ,...,K n ；
When one license plate image tu (0) needs to be scored, preprocessing the license plate image tu (0) into a license plate image with uniform size to obtain a license plate image tu (1);
sequentially adopting an effective scoring algorithm SF 1 ,SF 2 ,...,SF n Calculating the image definition scores of the license plate images tu (1), wherein the scoring results are respectively as follows: score 1 ,score 2 ,...,score n ；
The total image sharpness score for the license plate image tu (1) is calculated using:
wherein:
W i representing an efficient scoring algorithm SF i Is determined by the algorithm weight of (a);
score (max) is a preset score maximum;
score (min) is a preset score minimum.
Preferably, in the step 2, the license plate image effective scoring algorithm set is constructed, updated and expanded.
Preferably, in step 2.5, it is determined whether the image sharpness distribution map of the clear license plate image sample and the image sharpness distribution map of the blurred license plate image sample have a significant difference, specifically:
and the overlapping area of the area corresponding to the image definition distribution map of the clear license plate image sample and the area corresponding to the image definition distribution map of the blurred license plate image sample is smaller than an area setting threshold.
The scoring method for the license plate image non-reference quality analysis has the following advantages:
the scoring method solves the limitation that the full-reference image quality evaluation depends on a label image and the singleness of the non-reference image quality evaluation method, comprehensively considers the license plate image quality of various factors such as the image horizontal gradient, edge detection, gray variance, energy gradient, edge sharpness, secondary blurring, low-pass filtering, image structure similarity and the like, has more elements, and is more objective and accurate in evaluation.
Drawings
FIG. 1 is a flow chart of a scoring method for license plate image non-reference quality analysis;
FIG. 2 is an image sharpness distribution map of a sharp license plate image sample;
fig. 3 is an image sharpness distribution diagram of a blurred license plate image sample.
Detailed Description
In order to make the technical problems, technical schemes and beneficial effects solved by the invention more clear, the invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
In order to solve the problem that full-reference image quality evaluation depends on the limitation of a tag image (difficulty in application in a real scene) and singleness (low reliability) of non-reference image quality evaluation, the invention synthesizes various image evaluation methods, considers the image quality evaluation algorithms of a license plate such as image horizontal gradient, edge detection of a sobel operator and a laplace operator template, gray variance, energy gradient, edge sharpness, secondary blurring, low-pass filtering, image structure similarity and the like, realizes the scoring of the license plate image without reference quality analysis, does not depend on the tag image, and effectively improves the reliability of license plate image quality evaluation.
The invention provides a scoring method for no-reference quality analysis of license plate images, which is a scoring method for no-reference image quality analysis of comprehensive multiple factors, and referring to FIG. 1, the scoring method comprises the following steps:
step 1, obtaining a license plate image sample total set S;
step 1.1, obtaining m license plate images; preprocessing each license plate image, and processing each license plate image into a license plate image with uniform size as a license plate image sample; as an embodiment, the method for preprocessing each license plate image is as follows: and (3) performing perspective transformation and correction on the license plate image by utilizing the four-corner information of the license plate image, and scaling to 96 x 96 dimensions.
Obtaining m license plate image samples; each license plate image sample is provided with a sample label; the sample label comprises a clear license plate image label or a fuzzy license plate image label;
in the m Zhang Chepai image samples, m is included 1 Clear license plate image sample and m 2 A blurred license plate image sample is stretched;
for example, a total of 6000 license plate image samples, including 3000 clear license plate image samples and 3000 blurred license plate image samples;
step 1.2, m 1 The clear license plate image samples form a clear license plate image sample set S 1 The method comprises the steps of carrying out a first treatment on the surface of the Let m 2 The blurred license plate image samples form blurred license plate image sample set S 2 ；
Clear license plate image sample set S 1 And blurred license plate image sample set S 2 Composing license plate imagesA sample total set S;
step 2, constructing a license plate image effective scoring algorithm set, wherein the license plate image effective scoring algorithm set comprises n license plate image effective scoring algorithms, which are respectively expressed as: effective scoring algorithm SF 1 ,SF 2 ,...,SF n ；
The specific method comprises the following steps:
for each license plate image scoring algorithm G, identifying whether the license plate image scoring algorithm G is a license plate image effective scoring algorithm, if so, adding the license plate image scoring algorithm G into the license plate image effective scoring algorithm set;
wherein: the method is characterized in that whether the license plate image scoring algorithm G is a license plate image effective scoring algorithm is identified by adopting the following method:
step 2.1, respectively calculating clear license plate image sample sets S by adopting a license plate image scoring algorithm G 1 Image definition score of each clear license plate image sample in the license plate image is calculated to obtain m 1 Scoring the definition of each image;
step 2.2, for m 1 Counting the distribution of the individual image definition scores to obtain an image definition distribution map of a clear license plate image sample, wherein the abscissa is the image definition score and the ordinate is the sample number;
step 2.3, respectively calculating a fuzzy license plate image sample set S by adopting a license plate image scoring algorithm G 2 Image definition score of each blurred license plate image sample, thereby calculating m 2 Scoring the definition of each image;
step 2.4, for m 2 Counting the distribution of the individual image definition scores to obtain an image definition distribution map of a fuzzy license plate image sample with the abscissa being the image definition score and the ordinate being the sample number;
step 2.5, judging whether the image definition distribution map of the clear license plate image sample and the image definition distribution map of the fuzzy license plate image sample have obvious differences, if so, obtaining a conclusion that a license plate image scoring algorithm G is a license plate image effective scoring algorithm; otherwise, a conclusion that the license plate image scoring algorithm G is a license plate image invalid scoring algorithm is obtained;
in step 2.5, whether the image definition distribution map of the clear license plate image sample and the image definition distribution map of the blurred license plate image sample have obvious differences is judged, specifically:
and the overlapping area of the area corresponding to the image definition distribution map of the clear license plate image sample and the area corresponding to the image definition distribution map of the blurred license plate image sample is smaller than an area setting threshold. The smaller the overlapping area of the two groups of image definition distribution diagrams is, the better the algorithm performance is, and the license plate image effective scoring algorithm set can be preferentially included.
In addition, the license plate image effective scoring algorithm set has expandability, and the number of the incorporated algorithms can be increased or reduced according to the effectiveness of each algorithm in distinguishing the image definition.
As an embodiment, at present, a license plate image effective scoring algorithm set includes 9 license plate image effective scoring algorithms: gaussian_laplacian, brenner, blurnois, nrss, tenengard, laplacian, smad 2, eav, energy_gradient.
Through analyzing each license plate image effective scoring algorithm, an image definition distribution map of a clear license plate image sample and an image definition distribution map of a fuzzy license plate image sample are obtained, and through verification, obvious differences of the two image definition distribution maps can be observed. The 9 license plate image effective scoring algorithms have remarkable actual effect on distinguishing the image definition, and can be incorporated into a license plate image effective scoring algorithm set. As a specific example, as shown in fig. 2, an image sharpness distribution map of a sharp license plate image sample is shown; as shown in fig. 3, which is an image sharpness distribution diagram of a blurred license plate image sample, the quality score distribution situation of the algorithm on two sets of license plate image samples and the actual effect of distinguishing the image sharpness can be observed from fig. 2 and 3, and the license plate image valid scoring algorithm set can be judged and included.
Step 3, for each effective scoring algorithm SF i Wherein i=1, 2, n, determining a corresponding algorithm thresholdValue K i ：
Step 3.1, determining an effective scoring algorithm SF i Initial threshold K of (2) i (0):
SF is adopted by an effective scoring algorithm i Respectively calculating fuzzy license plate image sample sets S 2 The image definition score of each blurred license plate image sample is selected, and the image definition score with the lowest image definition score is used as an effective scoring algorithm SF i Initial threshold K of (2) i (0)；
Step 3.2, presetting a step rho;
step 3.3, from the initial threshold value K i (0) Starting, traversing by taking the step rho as an increment value to obtain a threshold meeting the following objective function as an effective scoring algorithm SF i Is the optimal algorithm threshold K of (2) i ：
Objective function:
the classification error rate does not exceed a set value epsilon, for example, the set value epsilon is 1%; and, from the blurred license plate image sample set S 2 Identifying the number E of the fuzzy license plate images to be the largest;
wherein:
the classification error rate is calculated by the following method:
1) For each license plate image sample in the license plate image sample total set S, representing the license plate image sample as a license plate image sample P;
2) SF is adopted by an effective scoring algorithm i Identifying a license plate image sample P to obtain a classification result; wherein, the classification result is: the license plate image sample P is a blurred license plate image sample or a clear license plate image sample;
specifically, an effective scoring algorithm SF is adopted i Calculating to obtain an image definition Score (P) of the license plate image sample P;
comparing whether the image sharpness Score (P) is greater than the threshold value K of the current traversal i (0) +xρ, where x is the number of steps traversed;
if the number is larger than the number, obtaining a classification result that the license plate image sample P is a clear license plate image sample; otherwise, obtaining a classification result that the license plate image sample P is a fuzzy license plate image sample;
3) Comparing the classification result of the license plate image sample P with a sample label of the license plate image sample P, and if the classification result is consistent with the sample label, indicating that the classification of the license plate image sample P is correct; otherwise, indicating that the classification of the license plate image sample P is wrong;
4) Therefore, for each license plate image sample in the license plate image sample total set S, a classification result is obtained, and the classification error rate is calculated by adopting the following formula:
classification error rate = number of samples with classification errors/total number of license plate image samples of the license plate image sample total set S;
from a blurred license plate image sample set S 2 In the method, the number E of the fuzzy license plate images is identified, and the method is adopted for calculating the number E of the fuzzy license plate images:
1) For a blurred license plate image sample set S 2 Each fuzzy license plate image sample in the system adopts an effective scoring algorithm SF i Identifying the obtained product to obtain a classification result; wherein, the classification result is: the blurred license plate image sample is a blurred license plate image sample or a clear license plate image sample;
2) For a blurred license plate image sample set S 2 Comprising m 2 The fuzzy license plate image samples are expanded, and the number of the fuzzy license plate image samples is counted to obtain a classification result, namely the number E of the fuzzy license plate images is identified;
step 4, thereby determining the effective scoring algorithm SF 1 ,SF 2 ,...,SF n The algorithm thresholds of (a) are respectively: k (K) 1 ,K 2 ,...,K n ；
When one license plate image tu (0) needs to be scored, preprocessing the license plate image tu (0) into a license plate image with uniform size to obtain a license plate image tu (1);
sequentially adopting an effective scoring algorithm SF 1 ,SF 2 ,...,SF n Calculating the image definition scores of the license plate images tu (1), wherein the scoring results are respectively as follows: score 1 ,score 2 ,...,score n ；
The total image sharpness score for the license plate image tu (1) is calculated using:
wherein:
W i representing an efficient scoring algorithm SF i Is determined by the algorithm weight of (a);
score (max) is a preset score maximum; for example, 99.9;
score (min) is a preset score minimum. For example, 39.9;
one specific embodiment is described below:
step 1, obtaining a license plate image sample total set S;
the total license plate image sample set S comprises 6000 license plate image samples in total, wherein the 6000 license plate image samples comprise 3000 clear license plate image samples and 3000 fuzzy license plate image samples;
wherein: each license plate image sample was 96 x 96 in size.
Step 2, constructing a license plate image effective scoring algorithm set which comprises 9 license plate image effective scoring algorithms, wherein the license plate image effective scoring algorithm set comprises the following steps of: gaussian_laplacian, brenner, blurnois, nrss, tenengard, laplacian, smad 2, eav, energy_gradient. The 9 license plate image effective scoring algorithms are non-reference image quality scoring algorithms.
For each license plate image effective scoring algorithm, counting scoring distribution of 3000 clear license plate image samples to obtain an image definition distribution map of the clear license plate image samples shown in fig. 2; counting the scoring distribution of 3000 fuzzy license plate image samples to obtain an image definition distribution map of the fuzzy license plate image samples shown in fig. 3; the overlapping area of the two sets of image sharpness profiles is small, indicating a significant difference.
And 3, for each effective scoring algorithm, determining a corresponding algorithm threshold.
Initial threshold setting: for each effective scoring algorithm, selecting the image definition score with the lowest image definition score as the initial threshold value of the effective scoring algorithm by scoring the image definition of 3000 fuzzy license plate image samples
Starting from an initial threshold value, gradually increasing the threshold value by taking 2 as a step length to obtain the threshold value meeting the following objective function as the optimal algorithm threshold value of the effective scoring algorithm:
objective function:
the classification error rate is not more than 1%; and identifying the maximum number of fuzzy license plate images from 3000 fuzzy license plate image samples;
in this embodiment, the optimal threshold set by the 9 effective scoring algorithms:
the thresholds for gaussian_laplacian, brenner, blast, nrss, tenengard, laplacian, smd2, eav, and energy_gradient are respectively:
K 1 ＝20,K 2 ＝25,K 3 ＝65,K 4 ＝75,K 5 ＝14,K 6 ＝12,K 7 ＝9,K 8 ＝40,K 9 ＝30。
step 4: scoring method test
Test dataset:
a clear license plate image sample set 1 (4153 sheets) and a slightly blurred license plate image sample set A (4153 sheets) form 4153 pair images, and a sample set 1A;
a clear license plate image sample set 1 (4153 sheets) and a moderate fuzzy license plate image sample set B (4153 sheets) form 4153 pair images, and a sample set 1B;
the clear license plate image sample set 2 (7701) and the severe blurred license plate image sample set C (7701) form 7701 pair images, and the license plate image sample set 2C.
The test indexes are as follows:
sample classification accuracy.
For example, in the sample set 1B, for 4153×2=8306 image samples, for each sample, scoring the sample by adopting any one effective scoring algorithm, and if the score is greater than a threshold value set by the effective scoring algorithm, obtaining a classification result that the sample is a clear license plate image sample; otherwise, obtaining a classification result that the sample is a fuzzy license plate image sample; through experiments, 3866 image samples are successfully identified as classification results of distinct license plate image samples; accuracy = 3866/4153 = 93.08% respectively
The classification accuracy results of each sample set are as follows:
step 5: when one license plate image tu (0) needs to be scored, preprocessing the license plate image tu (0) into a license plate image with uniform size to obtain a license plate image tu (1);
sequentially adopting an effective scoring algorithm SF 1 ,SF 2 ,...,SF 9 Calculating the image definition scores of the license plate images tu (1), wherein the scoring results are respectively as follows: score 1 ,score 2 ,...,score 9 ；
The total image sharpness score for the license plate image tu (1) is calculated using:
that is, only when the 9 effective scoring algorithms recognize that the license plate image tu (1) is a clear license plate image, the license plate image tu (1) is considered to be a clear license plate image, the total score of the evaluation does not exceed 99.9 score, and a smaller value of the weighted sum of the scores of the 99.9 and the algorithms is taken as the final score; otherwise, only one effective scoring algorithm recognizes that the license plate image tu (1) is a fuzzy license plate image, namely the license plate image tu (1) is considered as a fuzzy license plate image, the total score is evaluated to be not more than 39.9 score, and a smaller value of the weighted sum of the 39.9 score and the score of each algorithm is taken as a final score;
therefore, in the invention, a scoring system formed by a scoring method without reference quality analysis for license plate images is input: a license plate image with size (96, 96) in an actual scene; and (3) outputting: the sharpness score of the license plate image.
The scoring system combines an effective evaluation algorithm taking various factors into consideration, such as image horizontal gradient, edge detection, gray variance, energy gradient, edge sharpness, secondary blurring, low-pass filtering, image structure similarity and the like. The sharpness score of the image can be calculated by combining the comprehensive models of a plurality of effective evaluation algorithms. The higher the sharpness score, the better the representative image quality and the better the sharpness.
Compared with the prior art, the invention has the beneficial effects that:
(1) The scoring method for the license plate image without reference quality analysis has clear frame and definite input and output.
(2) The scoring method has expandability, and the number of the scoring algorithms to be incorporated can be increased or decreased according to the effectiveness of each scoring algorithm in distinguishing the image definition.
(3) The scoring method can be optimized, parameters are adjusted for each scoring algorithm, the set scoring algorithm threshold values are adjusted, and the scoring accuracy of the whole system can be optimized.
(4) The scoring method can be used for testing, outputting the score of each image by collecting clear and fuzzy license plate images and inputting the clear and fuzzy license plate images into a system, and finally calculating the classification accuracy.
(5) The scoring method solves the limitation that the full-reference image quality evaluation depends on a label image and the singleness of the non-reference image quality evaluation method, comprehensively considers the license plate image quality of various factors such as the image horizontal gradient, edge detection, gray variance, energy gradient, edge sharpness, secondary blurring, low-pass filtering, image structure similarity and the like, has more elements, and is more objective and accurate in evaluation.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which is also intended to be covered by the present invention.
Claims (3)
1. The scoring method for the license plate image non-reference quality analysis is characterized by comprising the following steps of:
step 1, obtaining a license plate image sample total set S;
step 1.1, obtaining m license plate images; preprocessing each license plate image, and processing each license plate image into a license plate image with uniform size as a license plate image sample;
obtaining m license plate image samples; each license plate image sample is provided with a sample label; the sample label comprises a clear license plate image label or a fuzzy license plate image label;
in the m Zhang Chepai image samples, m is included 1 Clear license plate image sample and m 2 A blurred license plate image sample is stretched;
step 1.2, m 1 The clear license plate image samples form a clear license plate image sample set S 1 The method comprises the steps of carrying out a first treatment on the surface of the Let m 2 The blurred license plate image samples form blurred license plate image sample set S 2 ；
Clear license plate image sample set S 1 And blurred license plate image sample set S 2 Forming a license plate image sample total set S;
step 2, constructing a license plate image effective scoring algorithm set, wherein the license plate image effective scoring algorithm set comprises n license plate image effective scoring algorithms, which are respectively expressed as: effective scoring algorithm SF 1 ,SF 2 ,...,SF n ；
The specific method comprises the following steps:
for each license plate image scoring algorithm G, identifying whether the license plate image scoring algorithm G is a license plate image effective scoring algorithm, if so, adding the license plate image scoring algorithm G into the license plate image effective scoring algorithm set;
wherein: the method is characterized in that whether the license plate image scoring algorithm G is a license plate image effective scoring algorithm is identified by adopting the following method:
step 2.1, respectively calculating clear license plate image sample sets S by adopting a license plate image scoring algorithm G 1 Image definition score of each clear license plate image sample in the license plate image is calculated to obtain m 1 Scoring the definition of each image;
step 2.2, for m 1 Distribution of individual image sharpness scoresCounting to obtain an image definition distribution map of a clear license plate image sample, wherein the abscissa is the image definition score, and the ordinate is the sample number;
step 2.3, respectively calculating a fuzzy license plate image sample set S by adopting a license plate image scoring algorithm G 2 Image definition score of each blurred license plate image sample, thereby calculating m 2 Scoring the definition of each image;
step 2.4, for m 2 Counting the distribution of the individual image definition scores to obtain an image definition distribution map of a fuzzy license plate image sample with the abscissa being the image definition score and the ordinate being the sample number;
step 2.5, judging whether the image definition distribution map of the clear license plate image sample and the image definition distribution map of the fuzzy license plate image sample have obvious differences, if so, obtaining a conclusion that a license plate image scoring algorithm G is a license plate image effective scoring algorithm; otherwise, a conclusion that the license plate image scoring algorithm G is a license plate image invalid scoring algorithm is obtained;
step 3, for each effective scoring algorithm SF i Wherein i=1, 2, n, determining a corresponding algorithm threshold K i ：
Step 3.1, determining an effective scoring algorithm SF i Initial threshold K of (2) i (0):
SF is adopted by an effective scoring algorithm i Respectively calculating fuzzy license plate image sample sets S 2 The image definition score of each blurred license plate image sample is selected, and the image definition score with the lowest image definition score is used as an effective scoring algorithm SF i Initial threshold K of (2) i (0)；
Step 3.2, presetting a step rho;
step 3.3, from the initial threshold value K i (0) Starting, traversing by taking the step rho as an increment value to obtain a threshold meeting the following objective function as an effective scoring algorithm SF i Is the optimal algorithm threshold K of (2) i ：
Objective function:
the classification error rate does not exceed the set value epsilon; and, from blurred license plate image samplesSet S 2 Identifying the number E of the fuzzy license plate images to be the largest;
wherein:
the classification error rate is calculated by the following method:
1) For each license plate image sample in the license plate image sample total set S, representing the license plate image sample as a license plate image sample P;
2) SF is adopted by an effective scoring algorithm i Identifying a license plate image sample P to obtain a classification result; wherein, the classification result is: the license plate image sample P is a blurred license plate image sample or a clear license plate image sample;
specifically, an effective scoring algorithm SF is adopted i Calculating to obtain an image definition Score (P) of the license plate image sample P;
comparing whether the image sharpness Score (P) is greater than the threshold value K of the current traversal i (0) +xρ, where x is the number of steps traversed;
if the number is larger than the number, obtaining a classification result that the license plate image sample P is a clear license plate image sample; otherwise, obtaining a classification result that the license plate image sample P is a fuzzy license plate image sample;
3) Comparing the classification result of the license plate image sample P with a sample label of the license plate image sample P, and if the classification result is consistent with the sample label, indicating that the classification of the license plate image sample P is correct; otherwise, indicating that the classification of the license plate image sample P is wrong;
4) Therefore, for each license plate image sample in the license plate image sample total set S, a classification result is obtained, and the classification error rate is calculated by adopting the following formula:
classification error rate = number of samples with classification errors/total number of license plate image samples of the license plate image sample total set S;
from a blurred license plate image sample set S 2 In the method, the number E of the fuzzy license plate images is identified, and the method is adopted for calculating the number E of the fuzzy license plate images:
1) For a blurred license plate image sample set S 2 Each fuzzy license plate image sample in the system adopts an effective scoring algorithm SF i Identifying the obtained product to obtain a classification result; wherein the classification junctionThe method comprises the following steps: the blurred license plate image sample is a blurred license plate image sample or a clear license plate image sample;
2) For a blurred license plate image sample set S 2 Comprising m 2 The fuzzy license plate image samples are expanded, and the number of the fuzzy license plate image samples is counted to obtain a classification result, namely the number E of the fuzzy license plate images is identified;
step 4, thereby determining the effective scoring algorithm SF 1 ,SF 2 ,...,SF n The algorithm thresholds of (a) are respectively: k (K) 1 ,K 2 ,...,K n ；
When one license plate image tu (0) needs to be scored, preprocessing the license plate image tu (0) into a license plate image with uniform size to obtain a license plate image tu (1);
sequentially adopting an effective scoring algorithm SF 1 ,SF 2 ,...,SF n Calculating the image definition scores of the license plate images tu (1), wherein the scoring results are respectively as follows: score 1 ,score 2 ,...,score n ；
The total image sharpness score for the license plate image tu (1) is calculated using:
wherein:
W i representing an efficient scoring algorithm SF i Is determined by the algorithm weight of (a);
score (max) is a preset score maximum;
score (min) is a preset score minimum.
2. The scoring method for license plate image non-reference quality analysis according to claim 1, wherein in step 2, the obtained license plate image effective scoring algorithm set is constructed, updated and expanded.
3. The method for scoring a license plate image without reference quality analysis according to claim 1, wherein in step 2.5, it is determined whether the image sharpness distribution map of the sharp license plate image sample and the image sharpness distribution map of the blurred license plate image sample have a significant difference, specifically:
and the overlapping area of the area corresponding to the image definition distribution map of the clear license plate image sample and the area corresponding to the image definition distribution map of the blurred license plate image sample is smaller than an area setting threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111107282.1A CN113763378B (en) | 2021-09-22 | 2021-09-22 | Scoring method for license plate image non-reference quality analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111107282.1A CN113763378B (en) | 2021-09-22 | 2021-09-22 | Scoring method for license plate image non-reference quality analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113763378A CN113763378A (en) | 2021-12-07 |
| CN113763378B true CN113763378B (en) | 2023-07-18 |
Family
ID=78796665
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111107282.1A Active CN113763378B (en) | 2021-09-22 | 2021-09-22 | Scoring method for license plate image non-reference quality analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113763378B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020016836A1 (en) * | 2018-07-18 | 2020-01-23 | Advenio Tecnosys Pvt. Ltd. | System and method for managing the quality of an image |
| CN111340041A (en) * | 2020-03-13 | 2020-06-26 | 安阳工学院 | License plate recognition method and device based on deep learning |
-
2021
- 2021-09-22 CN CN202111107282.1A patent/CN113763378B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020016836A1 (en) * | 2018-07-18 | 2020-01-23 | Advenio Tecnosys Pvt. Ltd. | System and method for managing the quality of an image |
| CN111340041A (en) * | 2020-03-13 | 2020-06-26 | 安阳工学院 | License plate recognition method and device based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 基于深度迁移学习的淘宝图像质量分类系统的设计与实现;徐浩;《cnki硕士论文电子期刊》;全文 * |
| 无参考的车牌自动识别图像质量评价方法;潘薇;吴;;科学技术与工程(第33期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113763378A (en) | 2021-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115861135B (en) | Image enhancement and recognition method applied to panoramic detection of box body | |
| JP7458328B2 (en) | Multi-sample whole-slide image processing via multi-resolution registration | |
| CN110569878B (en) | Photograph background similarity clustering method based on convolutional neural network and computer | |
| CN110148162B (en) | Heterogeneous image matching method based on composite operator | |
| CN116664559B (en) | Machine vision-based memory bank damage rapid detection method | |
| US20090010544A1 (en) | Method, apparatus, and program for detecting facial characteristic points | |
| CN110580510B (en) | Clustering result evaluation method and system | |
| CN108537787B (en) | Quality judgment method for face image | |
| CN110782445A (en) | No-reference image quality evaluation method and system | |
| CN116597432B (en) | License plate recognition system based on improved yolov5 algorithm | |
| CN111340041A (en) | License plate recognition method and device based on deep learning | |
| CN116129195A (en) | Image quality evaluation device, image quality evaluation method, electronic device, and storage medium | |
| CN112215780B (en) | Image evidence obtaining and resistance attack defending method based on class feature restoration fusion | |
| CN115393861B (en) | Method for accurately segmenting handwritten text | |
| CN110728214B (en) | Weak and small figure target detection method based on scale matching | |
| CN116524269A (en) | Visual recognition detection system | |
| CN115511856A (en) | Component defect detection method and device, electronic equipment and storage medium | |
| CN113763378B (en) | Scoring method for license plate image non-reference quality analysis | |
| CN112991159B (en) | Face illumination quality evaluation method, system, server and computer readable medium | |
| CN106446832B (en) | Video-based pedestrian real-time detection method | |
| CN117218627A (en) | Subway obstacle detection method and device based on standardized flow | |
| CN116843715A (en) | Multi-view collaborative image segmentation method and system based on deep learning | |
| CN110874547A (en) | Method and device for identifying object from video | |
| CN111382703B (en) | Finger vein recognition method based on secondary screening and score fusion | |
| CN112085683A (en) | Depth map reliability detection method in significance detection |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |