CN115511907B - Scratch detection method for LED screen - Google Patents
Scratch detection method for LED screen Download PDFInfo
- Publication number
- CN115511907B CN115511907B CN202211478913.5A CN202211478913A CN115511907B CN 115511907 B CN115511907 B CN 115511907B CN 202211478913 A CN202211478913 A CN 202211478913A CN 115511907 B CN115511907 B CN 115511907B
- Authority
- CN
- China
- Prior art keywords
- expansion
- point
- value
- pixel point
- starting point
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the field of image processing, in particular to a scratch detection method for an LED screen, which comprises the following steps: acquiring gradient values of all pixel points in a modulation image of the LED screen; obtaining the initial point probability of each pixel point according to the local gradient obvious degree and the overall gradient obvious degree of each pixel point; clustering the initial point probability of each pixel point to obtain each expansion initial point; acquiring each expansion area of each expansion starting point; obtaining the pre-estimated segmentation effect of each expansion area according to the starting point probability value and the linear degree of each pixel point in each expansion area of each expansion starting point so as to obtain the final expansion area of each expansion starting point; and respectively carrying out self-adaptive threshold segmentation on each final expansion region to obtain a scratch region of the LED screen. The invention can adaptively set the threshold segmentation window for the scratches at different positions, thereby improving the accuracy of scratch detection.
Description
Technical Field
The invention relates to the field of image processing, in particular to a scratch detection method for an LED screen.
Background
The scratch defect of the LED screen is one of the defects of scenes in the production process of the LED screen. In the production process of the LED screen, if the LED screen is scratched, the imaging effect of the LED display screen is poor, the LED screen is influenced to be watched, and the LED screen belongs to poor products. When the scratch of the existing LED screen is detected, the scratch of the LED screen is detected by adopting structured light. And obtaining a modulation image through the 3d structured light, and performing threshold segmentation after obtaining the modulation image to realize final smooth scratch defect detection. However, when the modulated image is subjected to threshold segmentation, the traditional local threshold segmentation method usually uses a threshold segmentation window with a fixed size, but due to the fact that scratch depths at different positions of an LED screen are different, the traditional method is difficult to be suitable for scratch detection with different depths, overexposure occurs inside a partial region, at this time, whether scratches exist in the overexposure region cannot be judged, and therefore the actual segmentation effect is poor.
Disclosure of Invention
The invention provides a scratch detection method for an LED screen, which aims to solve the existing problems.
The invention discloses a scratch detection method for an LED screen, which adopts the following technical scheme:
an embodiment of the present invention provides a scratch detection method for an LED screen, including the steps of:
acquiring a modulation image of the LED screen, and acquiring gradient values of all pixel points in the modulation image;
obtaining the local gradient obvious degree of each pixel point according to the gradient value of each pixel point and the maximum gradient value and the minimum gradient value in the preset local area of each pixel point; obtaining the integral gradient obvious degree of each pixel point according to the gradient value of each pixel point and the maximum gradient value and the minimum gradient value in the modulation image; obtaining the initial point probability of each pixel point according to the local gradient obvious degree and the overall gradient obvious degree of each pixel point; clustering the initial point probability of each pixel point to obtain each expansion initial point;
taking a preset local area of each expansion starting point as an initial expansion area of each expansion starting point; according to a preset expansion step length, carrying out area expansion on the initial expansion area of each expansion starting point to obtain each expansion area corresponding to each expansion starting point; obtaining a first index of each expansion area corresponding to each expansion starting point according to the starting point probability value of each pixel point in each expansion area corresponding to each expansion starting point; obtaining a second index of each expansion area corresponding to each expansion starting point according to the linear degree of each pixel point in each expansion area; obtaining the estimated segmentation effect of each expansion area according to the first index and the second index of each expansion area; taking the expansion area with the maximum pre-estimated segmentation effect in each expansion area as the final expansion area of each expansion starting point;
and performing self-adaptive threshold segmentation on each final expansion region respectively, and obtaining a scratch region of the LED screen according to a segmentation result.
Preferably, the method for obtaining the local gradient significance of each pixel point comprises:
and recording the difference between the gradient value of each pixel point and the minimum gradient value in the preset local area of each pixel point as a first local difference, recording the difference between the maximum gradient value and the minimum gradient value in the preset local area of each pixel point as a second local difference, and taking the ratio of the first local difference value to the second local difference value of each pixel point as the local gradient obvious degree of each pixel point.
Preferably, the obtaining of each expansion starting point is as follows:
performing secondary classification on the probability value of the starting point of each pixel point by using a K-means algorithm to obtain two clustering results; respectively calculating the average value of the initial point probability of each pixel point in the two clustering results, and recording the larger average value of the two obtained average values as a first average value and recording the smaller average value as a second average value;
calculating a difference value between the first mean value and the second mean value, when the difference value between the first mean value and the second mean value is larger than or equal to a preset threshold value, performing scratch detection, and taking each pixel point in a clustering result corresponding to the first mean value as each expansion starting point; otherwise, scratch detection is not needed, and no expansion starting point exists.
Preferably, the method for acquiring each expansion region corresponding to each expansion starting point comprises:
and taking the preset local area of each expansion starting point as an initial expansion area, increasing the length and width of the preset local area by one expansion step length each time the preset local area is expanded once, taking the obtained area as an expansion area of each expansion starting point, repeating the method until the expansion is completed to the whole modulation image, and obtaining each expansion area corresponding to each expansion starting point.
Preferably, the method for acquiring the first index of each expansion area corresponding to each expansion starting point is as follows:
obtaining the average value of the starting point probabilities corresponding to all the pixel points in each expansion area corresponding to each expansion starting point and the maximum value of the starting point probabilities corresponding to all the pixel points in each expansion area; and carrying out negative correlation mapping on the ratio of the average value to the maximum value, and taking the mapping result as a first index of each expansion area corresponding to each expansion starting point.
Preferably, the method for obtaining the linearity of each pixel point in each expansion region comprises:
recording an image formed by the initial point probability of each pixel point in the whole modulation image as a probability image, acquiring a Hessian matrix of each pixel point in each expansion area on the probability image, and obtaining two characteristic values of the Hessian matrix and two corresponding characteristic vectors; and obtaining a cosine distance value between the eigenvector corresponding to the maximum eigenvalue and the eigenvector corresponding to the minimum eigenvalue corresponding to each pixel point, and taking the obtained cosine distance value as the linearity degree of each pixel point.
Preferably, the method for acquiring the second index of each expansion area corresponding to each expansion starting point is as follows:
and obtaining the reference weight of each pixel point according to the difference value between the maximum value of the initial point probability of each expansion area corresponding to each expansion initial point and the initial point probability of each pixel point in each expansion area, and taking the average value of the result obtained by multiplying the reference weight of each pixel point in each expansion area by the linear degree of each pixel point as the second index of each expansion area corresponding to each expansion initial point.
The invention has the beneficial effects that: calculating the probability of each pixel point as the obvious scratch position, namely the initial point probability, according to the local gradient obvious degree of each pixel point in the local area and the overall gradient obvious degree of the pixel point in the whole modulation image, so that the accuracy of identifying the obvious scratch position is ensured, and the interference of local noise points is avoided; then obtain the expansion initial point according to the initial point probability of each pixel in the modulation image, and calculate the segmentation effect of pre-estimating of each expansion region that each expansion initial point corresponds, obtain the expansion region that each expansion initial point corresponds when segmentation effect is the best, thereby obtain the final expansion region of each expansion initial point, the regional expansion of the obvious position of different mar has been accomplished, the phenomenon that LED screen mar detection effect is relatively poor that has avoided leading to because of threshold value segmentation window size sets up improperly appears, thereby realize that the mar self-adaptation on different positions sets up the threshold value and cut apart the window, improve the accuracy that LED screen surface mar detected.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart illustrating the steps of a scratch detection method for an LED screen according to the present invention.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the scratch detection method for LED screen according to the present invention with reference to the accompanying drawings and preferred embodiments shows the following detailed descriptions of the specific implementation, structure, features and effects thereof. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The following describes a specific scheme of the scratch detection method for an LED screen in detail with reference to the accompanying drawings.
Referring to fig. 1, a flowchart illustrating steps of a scratch detection method for an LED screen according to an embodiment of the present invention is shown, where the method includes the following steps:
101: and acquiring a modulation image of the LED screen, and acquiring the gradient value of each pixel point in the modulation image.
When the scratch detection of the LED screen is carried out, the LED screen is fixed on a production line, the LED screen is polished by the 3d structured light equipment, the structured light image of the LED screen is collected by the camera matched with the 3d structured light equipment, a plurality of fringe patterns with the same phase shift on the LED screen are obtained, and all the fringe patterns are processed by using a modulation algorithm to obtain a modulation image of the LED screen. Then, denoising the modulation image by using a wavelet denoising algorithm, performing histogram equalization processing on the denoised modulation image by using a histogram equalization algorithm to complete contrast enhancement of the modulation image, and finally acquiring gradient values of all pixel points in the contrast-enhanced modulation image by using a Sobel operator.
102: obtaining the initial point probability of each pixel point according to the local gradient obvious degree and the overall gradient obvious degree of each pixel point; and clustering the initial point probability of each pixel point to obtain each expansion initial point.
Since the intensity of the structured light at the scratch is weakened when the modulated image is imaged and is determined according to the intensity of the structured light, the invention obtains the modulated image through the multiframe 3d structured light image for the scratch detection of the LED screen. And the positions with obvious scratches in the modulation image often have higher gradient values, namely, the gradient of the surface of the LED screen is no longer uniform. In order to segment the exact scratch position in the modulated image, the segmentation may be performed by a local threshold segmentation algorithm. However, if the window size is not properly selected in the local threshold segmentation, some of the windows must be divided into defects in the segmentation, resulting in poor segmentation. The present solution is used for local area dilation by pre-selecting the starting point of the local area. The starting point of the scratch is certainly a pixel point with a large gradient value in the local area range, but the starting point is judged only according to the local large gradient value, and the scratch is easily interfered by local noise. Therefore, the gradient significance of each pixel point in the local area and the whole modulation graph is combined, each pixel point is the position with obvious scratches in the local area, and then the area expansion is carried out according to the positions with obvious scratches, so that each scratch area in the modulation graph is determined, therefore, the positions with obvious scratches are also called as the expansion starting points, so that:
presetting a local area as one
The size of the sliding window of (1) can be adjusted by an implementer according to a specific implementation scene, wherein N =11 is set in the invention, and each pixel point is taken as the central point of the sliding window to obtain a preset local area corresponding to each pixel point; obtaining the maximum gradient value and the minimum gradient value of all pixel points in a preset local area of each pixel point, comparing the difference value between the gradient value and the minimum gradient value of each pixel point with the difference value between the maximum gradient value and the minimum gradient value, and taking the obtained ratio as the local gradient obvious degree of each pixel point; similarly, the difference between the gradient value of each pixel point and the minimum gradient value in the whole modulation image is compared with the difference between the maximum gradient value and the minimum gradient value in the whole modulation image, and the obtained ratio is used as the integral gradient obvious degree of each pixel point; taking the product of the local gradient obvious degree and the integral gradient obvious degree of each pixel point as the starting point probability of each pixel point, wherein the starting point probability of the ith pixel point in the modulation image is->
Can be expressed as:
in the formula (I), the compound is shown in the specification,
the gradient value of the ith pixel point is obtained; />
The maximum gradient value and the minimum gradient value of all pixel points contained in the local area of the ith pixel point are respectively; />
The maximum gradient value and the minimum gradient value of all pixel points contained in the whole modulation image are respectively.
because some pixel points which are not the obvious scratch position are easily judged to be the obvious scratch position by mistake only according to the gradient value in the local area of each pixel point, for example, for one pixel point, no scratch exists in the local area corresponding to the pixel point, namely, the gradient values of all the pixel points in the local area of the pixel point are smaller, but because the gradient value of the pixel point is slightly higher than the gradient values of the pixel points in other positions, if the pixel point is also the obvious scratch position only according to the obvious local gradient degree, namely, the judgment result obtained only according to the obvious local gradient degree is not accurate, whether the pixel point is an expansion starting point or not needs to be comprehensively judged by combining the obvious integral gradient degree of each pixel point in the whole modulation image, thereby eliminating the interference of local noise points
The integral gradient obvious degree of the ith pixel point is expressed, the larger the value is, the larger the integral gradient obvious degree of the pixel point is, the higher the probability that the pixel point is at the obvious position of the scratch is, and the higher the probability corresponding to the expansion starting point is.
Processing each pixel point by using the method to obtain the initial point probability of each pixel point, and performing two on the initial point probability of each pixel point in the modulated image by using a K-means algorithmClassifying to obtain two clustering results, wherein the local gradient obvious degree and the whole gradient obvious degree of the scratch obvious position are both large, namely the probability value of the starting point of the expansion starting point is high, the average value of the starting point probabilities of all the pixel points in the two clustering results is respectively calculated, the larger average value in the two obtained average values is recorded as a first average value, the smaller average value is recorded as a second average value, the scratch possibly does not exist in an LED screen, the probability of the starting point of each pixel point in the whole modulation image is smaller at the moment, the difference value between the corresponding first average value and the corresponding second average value is also smaller, and therefore the preset threshold value is smaller
When the difference between the first mean value and the second mean value is less than ^ or ^>
When the probability difference of the starting point between the two clustering results is small, the fact that no scratch exists on the surface of the LED screen at the moment is shown; and if not, determining that scratches exist on the surface of the LED screen, and taking each pixel point in the clustering result corresponding to the first average value as each expansion starting point.
103: acquiring each expansion area corresponding to each expansion starting point; and obtaining the pre-estimated segmentation effect of each expansion area corresponding to each expansion starting point according to the starting point probability value and the linear degree of each pixel point in each expansion area, and further obtaining the final expansion area of each expansion starting point.
The scratch position is characterized in that the gradient value of the scratch position in a local area is greatly different from the gradient value of peripheral pixels, and if the whole modulation image is subjected to threshold segmentation, the phenomenon of overexposure in the local area can be caused, so that the invention expects that the threshold segmentation can be carried out in the local area, and the detection of scratches at different positions is realized. The selection of the size of the window of the local area often affects the segmentation effect of the scratch area, so the invention needs to determine the size of the local area with the best segmentation effect on each scratch, the preset local area of each expansion starting point is used as an initial expansion area, the expansion is carried out to the periphery according to the expansion step length, the expansion step length is set, the size of the expansion step length can be adjusted by an implementer according to a specific real-time scene, the expansion step length is set to be 1, namely, the length and the width of the window of the local area corresponding to each expansion starting point are respectively added with 1 to obtain a corresponding expansion area during each expansion; until the whole modulation image is expanded, obtaining each expansion area corresponding to each expansion starting point;
when the local region threshold segmentation is carried out, the Otsu threshold segmentation algorithm is selected to carry out the self-adaptive threshold segmentation on each expansion region, so that in the process of judging whether the segmentation effect of each expansion region is the best expansion region or not, the judgment can be carried out according to the Otsu threshold segmentation effect, finally, the expansion region with the best segmentation effect in each expansion region corresponding to each expansion starting point is used as the final expansion region of each expansion starting point, and the local region expansion of each expansion starting point is completed. Wherein the predicted segmentation effect of the kth expansion region for the jth expansion starting point
Can be expressed as:
in the formula (I), the compound is shown in the specification,
the estimated segmentation effect of the kth expansion area representing the jth expansion starting point; />
Representing the average value of the starting point probabilities of all the pixel points in the kth expansion region; />
Expressing the maximum value of the starting point probability of all the pixel points in the kth expansion area; />
Representing the total number of pixel points contained in the kth expansion region; />
Representing the probability of the starting point of the r-th pixel point in the k-th expansion area; />
Represents the linear degree of the r-th pixel point in the k-th expansion area and is combined with the linear degree of the r-th pixel point in the k-th expansion area>
Is an exponential function with a natural constant as the base.
Because the probability of the starting point reflects the possibility that one pixel point is the obvious position of the scratch, the more difference between the scratch and the background is, namely, the more background is contained, the more obvious the scratch is, and the better the segmentation effect at the moment is. The probability values of the starting points corresponding to the pixel points in the background area are all smaller, so that the more the number of the background pixel points is, the more the starting points are, the obtained starting points are
The smaller, i.e.. Sup>
The number of background pixel points can be reflected; since the probability value of the starting point of a marked position is higher, in the case of a thresholding operation, then>
The smaller the value of (A), the more obvious the scratch is divided, the better the corresponding division effect, and similarly, when ^ is greater or less than>
The bigger, the more the pixel quantity of the obvious position of scratch in the corresponding expansion region is, and at this moment, because the segmentation difficulty between these points is bigger, the obtained segmentation effect is worse. Therefore, the present invention will->
Recording as the first index of the estimated segmentation effect and comparing ^ with ^ or ^ based on the first index>
Make a negative correlation mapping, i.e. when>
The smaller the corresponding segmentation effect is.
However, it is not limited to
If too small, due to some noise in the image, adaptive threshold segmentation of dilated regions results in a lower segmentation threshold, which segments the less homogeneous region of the modulated image, and therefore ^ based on>
If the pixel distribution is too small, the pixel distribution does not represent a good segmentation effect, so that it needs to be judged that the distribution of the pixel points belonging to the obvious position of the scratch in the expansion area of the jth expansion starting point should be relatively dispersed and linear, and cannot be concentrated into blocks.
So as to calculate the probability value of the starting point of each pixel point in the kth expansion area of the jth expansion starting point and
the smaller the difference is, the greater the probability that the pixel point is the obvious position of the scratch is, so the construction method of the invention is used for determining whether the scratch is obvious or not>
The negative correlation mapping model takes the mapping result as the r pixel point in the k expansion regionIn the present invention, for
Constructed negative correlation model is ^ er>
(ii) a At this time, if the r-th pixel point is more approximate to linearity, the scratch segmentation effect expressed by the r-th pixel point is better.
The image formed by the starting point probabilities of all the pixel points in the whole modulation image is called a probability image, wherein when the linearity of the r-th pixel point is represented, if the linearity is adopted, the variation of the starting point probabilities of the peripheral pixel points of the r-th pixel point in the probability image is smaller along the scratch direction, and the variation of the starting point probabilities of the peripheral pixel points of the r-th pixel point in the probability image is larger along the vertical direction of the scratch. Therefore, a hessian matrix of the r-th pixel point on the probability image is obtained, wherein the hessian matrix is a 2 x 2 matrix, represents a second derivative of the r-th pixel point on the probability image, can obtain two eigenvalues of the hessian matrix and two corresponding eigenvectors, and then obtains a cosine distance value between the eigenvector corresponding to the maximum eigenvalue and the eigenvector corresponding to the minimum eigenvalue of the r-th pixel point
And the linear degree of the r-th pixel point is expressed. Due to->
The probability image is represented according to the probability value change of the periphery of the corresponding pixel point of the probability image, wherein the probability value change is smaller and the change of the probability value of the starting point in a larger direction is represented, so that the larger the value is, the more linear the probability value is, and the stronger the corresponding linear degree is; i.e. is>
The smaller the>
The larger the segmentation effect is, the better the segmentation effect is, otherwise, the worse the segmentation effect is; the invention is to
A second index of a k-th dilated area as a jth dilation start point. And finally, obtaining the estimated segmentation effect of the jth expansion area according to the first index and the second index of each expansion area of the jth expansion starting point.
However, as the local area expands, the expanded area gradually contains more scratches, and the different scratches interfere with each other, so that the threshold segmentation effect of different expanded areas corresponding to the same expansion starting point is reduced, and therefore, the expanded area with the largest estimated segmentation result of each expanded area corresponding to the jth expansion starting point is used as the final expanded area of the jth expansion starting point;
repeating the above method to obtain the final expansion area of each expansion starting point.
Particularly, when the final expansion areas are overlapped, the overlapped area is divided into the final expansion area with the minimum estimated segmentation effect in the final expansion areas, so that the segmentation effect of the final expansion area with poor estimated segmentation effect is improved, and the segmentation effect of the partial area with smaller estimated segmentation effect value on the scratch is ensured. For the areas that are not expanded, because these areas have no expansion starting point and have little meaning to participate in the segmentation, the parts of the areas that are not expanded can be processed without being processed, and the default is the traceless scratch area.
104: and respectively carrying out self-adaptive threshold segmentation on each final expansion region to obtain a scratch region of the LED screen.
And (3) performing threshold segmentation of the self-adaptive local area on each expansion area obtained in the step (103) by using an Otsu threshold segmentation algorithm to obtain a self-adaptive threshold segmentation result of each expansion area. And at the moment, the threshold segmentation result of each expansion area is the scratch detection result of the LED screen. And superposing the threshold segmentation results of all the expansion areas, namely extracting pixel points of which the result is 1 after all the threshold segmentation in the modulation image, wherein the pixel points are scratch areas in the LED screen.
Through the steps, the detection of the scratches on the surface of the LED screen is completed.
According to the method, the probability that each pixel point is at the obvious scratch position, namely the initial point probability, is calculated according to the local gradient obvious degree of each pixel point in the local area and the overall gradient obvious degree of the pixel point in the whole modulation image, so that the accuracy of identifying the obvious scratch position is ensured, and the interference of local noise points is avoided; then obtain the expansion initial point according to the initial point probability of each pixel in the modulation image, and calculate each expansion initial point to each expansion regional prediction segmentation effect that, obtain the expansion region that each expansion initial point segmentation effect corresponds when best, thereby obtain the final expansion region of each expansion initial point, the regional expansion of the obvious position of different mar has been accomplished, the phenomenon that LED screen mar detection effect is relatively poor that has avoided leading to because of threshold value segmentation window size sets up improperly appears, thereby realize that the mar self-adaptation on different positions sets up the threshold value and cut apart the window, improve the accuracy that LED screen surface mar detected.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (6)
1. A scratch detection method for an LED screen, characterized in that the method comprises the following steps:
acquiring a modulation image of the LED screen, and acquiring gradient values of all pixel points in the modulation image;
obtaining the local gradient obvious degree of each pixel point according to the gradient value of each pixel point and the maximum gradient value and the minimum gradient value in the preset local area of each pixel point; obtaining the integral gradient obvious degree of each pixel point according to the gradient value of each pixel point and the maximum gradient value and the minimum gradient value in the modulation image; obtaining the initial point probability of each pixel point according to the local gradient obvious degree and the overall gradient obvious degree of each pixel point; clustering the initial point probability of each pixel point to obtain each expansion initial point;
taking a preset local area of each expansion starting point as an initial expansion area of each expansion starting point; according to a preset expansion step length, carrying out area expansion on the initial expansion area of each expansion starting point to obtain each expansion area corresponding to each expansion starting point; obtaining a first index of each expansion area corresponding to each expansion starting point according to the starting point probability value of each pixel point in each expansion area corresponding to each expansion starting point; obtaining a second index of each expansion area corresponding to each expansion starting point according to the linear degree of each pixel point in each expansion area; obtaining the estimated segmentation effect of each expansion area according to the first index and the second index of each expansion area; taking the expansion area with the maximum pre-estimated segmentation effect in each expansion area as the final expansion area of each expansion starting point;
respectively carrying out self-adaptive threshold segmentation on each final expansion region, and obtaining a scratch region of the LED screen according to a segmentation result;
the method for acquiring the linearity degree of each pixel point in each expansion area comprises the following steps:
recording an image formed by the initial point probability of each pixel point in the whole modulation image as a probability image, acquiring a Hessian matrix of each pixel point in each expansion area on the probability image, and obtaining two characteristic values of the Hessian matrix and two corresponding characteristic vectors; and obtaining a cosine distance value between the eigenvector corresponding to the maximum eigenvalue and the eigenvector corresponding to the minimum eigenvalue corresponding to each pixel point, and taking the obtained cosine distance value as the linearity degree of each pixel point.
2. The scratch detection method for the LED screen according to claim 1, wherein the method for obtaining the local gradient significance of each pixel point comprises:
and recording the difference between the gradient value of each pixel point and the minimum gradient value in the preset local area of each pixel point as a first local difference, recording the difference between the maximum gradient value and the minimum gradient value in the preset local area of each pixel point as a second local difference, and taking the ratio of the first local difference value to the second local difference value of each pixel point as the local gradient obvious degree of each pixel point.
3. The scratch detection method for an LED screen according to claim 1, wherein the obtaining of the respective expansion starting points is:
performing secondary classification on the probability value of the starting point of each pixel point by using a K-means algorithm to obtain two clustering results; respectively calculating the average value of the probability of the starting point of each pixel point in the two clustering results, and recording the larger average value as a first average value and recording the smaller average value as a second average value in the two obtained average values;
calculating a difference value between the first mean value and the second mean value, when the difference value between the first mean value and the second mean value is larger than or equal to a preset threshold value, performing scratch detection, and taking each pixel point in a clustering result corresponding to the first mean value as each expansion starting point; otherwise, scratch detection is not needed, and an expansion starting point does not exist.
4. The scratch detection method for the LED screen according to claim 1, wherein the method for obtaining each expansion area corresponding to each expansion starting point is as follows:
and taking the preset local area of each expansion starting point as an initial expansion area, increasing an expansion step length for the length and the width of the preset local area every time the preset local area is expanded once, taking the obtained area as an expansion area of each expansion starting point, and repeating the method until the expansion is finished until the whole modulation image, so as to obtain each expansion area corresponding to each expansion starting point.
5. The scratch detection method for the LED screen according to claim 1, wherein the first index of each expansion area corresponding to each expansion starting point is obtained by:
acquiring the average value of the initial point probabilities corresponding to all the pixel points in each expansion area corresponding to each expansion initial point and the maximum value of the initial point probabilities corresponding to all the pixel points in each expansion area; and carrying out negative correlation mapping on the ratio of the average value to the maximum value, and taking the mapping result as a first index of each expansion area corresponding to each expansion starting point.
6. The scratch detection method for the LED screen according to claim 1, wherein the second index of each expansion area corresponding to each expansion starting point is obtained by:
and obtaining the reference weight of each pixel point according to the difference value between the maximum value of the initial point probability of each expansion area corresponding to each expansion initial point and the initial point probability of each pixel point in each expansion area, and taking the average value of the result obtained by multiplying the reference weight of each pixel point in each expansion area by the linear degree of each pixel point as the second index of each expansion area corresponding to each expansion initial point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211478913.5A CN115511907B (en) | 2022-11-24 | 2022-11-24 | Scratch detection method for LED screen |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211478913.5A CN115511907B (en) | 2022-11-24 | 2022-11-24 | Scratch detection method for LED screen |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115511907A CN115511907A (en) | 2022-12-23 |
| CN115511907B true CN115511907B (en) | 2023-03-24 |
Family
ID=84513878
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211478913.5A Active CN115511907B (en) | 2022-11-24 | 2022-11-24 | Scratch detection method for LED screen |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115511907B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117252877B (en) * | 2023-11-17 | 2024-02-02 | 济南界龙科技有限公司 | Diode lead frame quality detection method based on image characteristics |
| CN117314899B (en) * | 2023-11-28 | 2024-03-08 | 深圳市烯碳复合材料有限公司 | Carbon fiber plate quality detection method based on image characteristics |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113781402A (en) * | 2021-08-19 | 2021-12-10 | 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） | Method and device for detecting chip surface scratch defects and computer equipment |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325716B (en) * | 2020-01-21 | 2023-09-01 | 上海万物新生环保科技集团有限公司 | Screen scratch and fragmentation detection method and equipment |
| CN111444921A (en) * | 2020-03-25 | 2020-07-24 | 浙江华睿科技有限公司 | Scratch defect detection method and device, computing equipment and storage medium |
| CN111612757B (en) * | 2020-05-18 | 2023-07-18 | 苏州精濑光电有限公司 | Screen crack detection method, device, equipment and storage medium |
| CN111815630B (en) * | 2020-08-28 | 2020-12-15 | 歌尔股份有限公司 | Defect detection method and device for LCD screen |
| CN112348773B (en) * | 2020-09-28 | 2022-09-23 | 歌尔股份有限公司 | Screen defect detection method and device and electronic equipment |
-
2022
- 2022-11-24 CN CN202211478913.5A patent/CN115511907B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113781402A (en) * | 2021-08-19 | 2021-12-10 | 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） | Method and device for detecting chip surface scratch defects and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115511907A (en) | 2022-12-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115511907B (en) | Scratch detection method for LED screen | |
| CN113781402B (en) | Method and device for detecting scratch defects on chip surface and computer equipment | |
| CN107507208B (en) | Image feature point extraction method based on curvature estimation on contour | |
| CN115619793B (en) | Power adapter appearance quality detection method based on computer vision | |
| CN109978848B (en) | Method for detecting hard exudation in fundus image based on multi-light-source color constancy model | |
| CN115049835A (en) | Data preprocessing method based on die-casting die defect identification | |
| CN115829883A (en) | Surface image denoising method for dissimilar metal structural member | |
| CN116310360A (en) | Reactor surface defect detection method | |
| CN116168026A (en) | Water quality detection method and system based on computer vision | |
| WO2021109697A1 (en) | Character segmentation method and apparatus, and computer-readable storage medium | |
| CN116740070A (en) | Plastic pipeline appearance defect detection method based on machine vision | |
| CN113191979B (en) | Non-local mean denoising method for partitioned SAR (synthetic aperture radar) image | |
| CN116883408B (en) | Integrating instrument shell defect detection method based on artificial intelligence | |
| CN111738931B (en) | Shadow removal algorithm for aerial image of photovoltaic array unmanned aerial vehicle | |
| CN117557820B (en) | Quantum dot optical film damage detection method and system based on machine vision | |
| CN117541582B (en) | IGBT insulation quality detection method for high-frequency converter | |
| CN117422712A (en) | Plastic master batch visual detection method and system based on image filtering processing | |
| CN113837198A (en) | Improved self-adaptive threshold Canny edge detection method based on three-dimensional block matching | |
| CN112529853A (en) | Method and device for detecting damage of netting of underwater aquaculture net cage | |
| CN110298835B (en) | Leather surface damage detection method, system and related device | |
| CN115346126A (en) | Side slope crack identification method | |
| CN117173180B (en) | Intelligent steel pipe defect detection method based on image enhancement | |
| CN115994870B (en) | Image processing method for enhancing denoising | |
| CN115100510B (en) | Tire wear degree identification method | |
| CN115660990A (en) | Endoscope image mirror reflection detection and restoration method based on brightness classification |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230525 Address after: 215637 east side of Guotai North Road, yangshe Town, Zhangjiagang City, Suzhou City, Jiangsu Province Patentee after: SUZHOU KINGLIGHT OPTOELECTRONICS CO.,LTD. Address before: 518100 Runheng Industrial Factory 501, Fuyuan 1st Road, Heping Community, Fuhai Street, Bao'an District, Shenzhen, Guangdong Patentee before: SHENZHEN JINGTAI Co.,Ltd. |