Image enhancement method for VCSEL epitaxial wafer detection

Abstract

The invention belongs to the technical field of image enhancement, and particularly relates to an image enhancement method for VCSEL (Vertical Cavity Surface Emitting Laser) epitaxial wafer detection, which comprises the following steps of: analyzing local and global gray information and structure tensor characteristics of pixel points, and calculating a defect saliency value of each pixel point to quantify the possibility of the pixel point as a defect; interference of a stripe structure in an epitaxial wafer image is weakened; a weighted histogram is constructed based on the defect saliency value, so that the defect pixel occupies a higher weight in the histogram; and determining an optimal segmentation threshold value by searching an energy median point of the weighted histogram, and finally performing BBHE enhancement on the image by using the segmentation threshold value. According to the method, the contrast ratio of the tiny defects can be remarkably improved, a high-quality image is provided for subsequent epitaxial wafer detection, and the detection accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology. More specifically, this invention relates to an image enhancement method for VCSEL epitaxial wafer detection. Background Technology

[0002] Vertical-cavity surface-emitting lasers (VCSELs) are core components in cutting-edge technologies such as optical communication, 3D sensing, and lidar. Their performance and reliability directly depend on the manufacturing quality of their underlying epitaxial wafers. During epitaxial wafer growth and subsequent processing, various microscopic defects, such as dislocations and black spots, are inevitably introduced. These defects are tiny in size and vary in shape, severely impacting the yield and long-term stability of VCSEL chips. Therefore, precise and efficient fully automated optical inspection of epitaxial wafers is crucial.

[0003] Currently, machine vision-based automated optical inspection systems are the mainstream technology for epitaxial wafer inspection. The inspection system acquires images of the epitaxial wafer surface and then uses image processing algorithms to identify potential defects. Image enhancement, as the first step in defect identification, directly determines the effectiveness of subsequent inspections. Bi-Histogram Equalization (BBHE) is a commonly used image contrast enhancement technique. This technique typically uses the image's gray-level mean as a dividing point and modulates the gray-level histogram... Figure 1 The image is divided into two sub-histograms, and then the two sub-histograms are subjected to equalization processing to improve the overall visual effect of the image.

[0004] However, defects in VCSEL epitaxial wafer images are extremely small, accounting for a very low percentage of pixels. The grayscale information of these minor defects has a negligible impact on the global grayscale mean. Therefore, using the mean as the segmentation point for the BBHE algorithm often groups these critical defect pixels together with background pixels into the same sub-histogram. When equalizing this sub-histogram, which contains a large amount of background information, the limited grayscale range is mainly allocated to the dominant background pixels, while the contrast enhancement effect on the truly needing defect pixels is limited. This results in the enhanced defects still not being significant enough, leading to missed detection of epitaxial wafer defects in subsequent defect identification steps. Summary of the Invention

[0005] To address the technical problem that the BBHE algorithm, which uses the gray-level mean of the image for gray-level histogram segmentation, has limited effect on enhancing epitaxial wafer images, this invention provides an image enhancement method for VCSEL epitaxial wafer detection. The method includes: taking any pixel in the image to be enhanced as the target pixel, obtaining a first discrimination index of the target pixel based on its local standard deviation, global standard deviation, and the difference between the gray-level value of the target pixel and its local mean; the local standard deviation is the standard deviation of the gray-level values ​​of the target pixel's neighboring pixels; the global standard deviation is the standard deviation of the gray-level values ​​of all pixels in the image to be enhanced; the local mean is the mean of the gray-level values ​​of the target pixel's neighboring pixels; and obtaining a first discrimination index of the target pixel based on its first standard deviation, global standard deviation, and the difference between the gray-level value of the target pixel and its local mean. The second discrimination of the target pixel is obtained by comparing the difference between the discrimination and the mean of the first discrimination of pixels in the row where the target pixel is located, and the first discrimination of the target pixel. The defect significance value of the target pixel is obtained based on the difference between two feature values ​​obtained from the target pixel's structure tensor and the second discrimination. The weighted values ​​of all pixels corresponding to each gray level in the image to be enhanced are accumulated, and a weighted histogram is obtained based on the accumulation result. The weighted values ​​are positively correlated with the defect significance value. The cumulative sum of the column heights corresponding to all gray levels between each gray level and the first gray level in the weighted histogram is obtained, and a segmentation threshold is obtained based on the cumulative sum of the column heights corresponding to the gray levels. The BBHE algorithm is used to enhance the image to be enhanced based on the segmentation threshold.

[0006] This invention obtains defect saliency values ​​that can effectively distinguish defects, background, and stripe interference by comprehensively analyzing the local and global grayscale information and structure tensor of pixels. A weighted histogram is then constructed based on these defect saliency values, allowing statistically minority defect pixels to receive higher weights in the histogram, thereby strengthening their impact on image enhancement. Finally, the optimal segmentation threshold is determined by finding the median energy point of the weighted histogram. This ensures that the segmentation threshold is based on the true defect saliency information in the image, rather than a simple pixel count, solving the problem of defect information being obscured by the background in traditional methods. This enables the image enhancement algorithm to significantly improve the contrast of minute defects, providing a high-quality image foundation for subsequent defect detection.

[0007] Preferably, the first discrimination index satisfies the following relationship: In the formula, For the first image to be enhanced The first distinguishing factor of each pixel For the first in the image to be enhanced The grayscale value of each pixel For the first in the image to be enhanced Local mean of each pixel For the first in the image to be enhanced Local standard deviation of each pixel The global standard deviation of the image to be enhanced. It is a linear normalization function.

[0008] This invention combines the grayscale difference between a pixel and its local mean and the ratio of the local standard deviation to the global standard deviation, which can more comprehensively quantify the degree of abnormality of a pixel in its neighborhood. It integrates the prominence of local details and the prominence of details in the global context, thus more accurately identifying potential defect areas than single feature judgment, making image enhancement more targeted.

[0009] Preferably, the second discrimination index satisfies the following relationship: In the formula, For the first in the image to be enhanced The second distinguishing factor of each pixel For the first in the image to be enhanced The first distinguishing factor of each pixel For the first in the image to be enhanced The mean of the first discrimination of all pixels in the row containing the given pixel. This is a correction factor.

[0010] This invention introduces the average first discrimination value of the row containing the pixel to correct the first discrimination value. Considering that VCSEL epitaxial film images generally have lateral stripe interference, these stripes can also produce high first discrimination values ​​locally, which can be easily confused with defects. By comparing the first discrimination value of a single pixel with the average first discrimination value of its row, the interference of this regular stripe structure on the image enhancement process can be effectively suppressed, reducing stripe interference and improving the targeting of image enhancement.

[0011] Preferably, the significant value of the defect satisfies the following relationship: In the formula, For the first in the image to be enhanced The significant value of defects per pixel, For the first in the image to be enhanced The second distinguishing factor of each pixel For the first in the image to be enhanced The larger of the two feature values ​​of a pixel. For the first in the image to be enhanced The smaller of the two feature values ​​of a pixel To prevent division by zero parameters.

[0012] This invention integrates a second discrimination factor and two eigenvalues ​​obtained through the structural tensor. It utilizes the difference in eigenvalues ​​of the structural tensor to characterize the directional characteristics of grayscale changes. By combining this characteristic with the second discrimination factor, it is possible to further distinguish between irregularly shaped defects and stripe structures with obvious directions, making the enhancement of defect locations more targeted and resulting in better image enhancement effects.

[0013] Preferably, obtaining the weighted histogram based on the accumulated results includes: sorting the accumulated weighted values ​​of all pixels corresponding to each gray level in ascending order of gray level to obtain the weighted histogram.

[0014] Preferably, the method for obtaining the weighted value includes: presetting a histogram weighting factor, and adding 1 to the result of multiplying the defect significance value by the histogram weighting factor to obtain the weighted value.

[0015] Preferably, the column height of each gray level in the weighted histogram satisfies the following relationship: In the formula, For the weighted histogram, the first A column height of one gray level, The grayscale value in the image to be enhanced is The The significant value of defects per pixel, The grayscale value in the image to be enhanced is The number of pixels, This is the weighting factor for the histogram.

[0016] Preferably, the step of obtaining the segmentation threshold based on the cumulative sum of the column heights corresponding to gray levels includes: obtaining the sum of the column heights of all gray levels in the weighted histogram, and taking half of the sum of the column heights of all gray levels as the median of significant energy; obtaining the cumulative sum of the column heights corresponding to all gray levels between each gray level and the first gray level in the weighted histogram, taking the difference between the cumulative sum and the median of significant energy as the energy difference, and taking the value corresponding to the gray level with the smallest energy difference as the segmentation threshold.

[0017] Preferably, the energy difference satisfies the following relationship: In the formula, For the weighted histogram, the first The energy difference of each gray level For the weighted histogram, the first A column height of one gray level, It represents a significant energy median.

[0018] This invention determines the segmentation threshold by using the median of significant energy, unlike existing technologies that use the mean or median of gray levels. By finding a gray level that evenly divides the total energy of the weighted histogram as the threshold, it ensures that the two sub-histograms after segmentation are balanced in terms of defect saliency. This allows subsequent BBHE enhancement to more specifically improve the contrast of defects and the background.

[0019] Preferably, the step of using the BBHE algorithm to enhance the image to be enhanced based on the segmentation threshold includes: segmenting the gray-level histogram of the image to be enhanced using the segmentation threshold, dividing the gray-level histogram into two parts, and then using the existing technology of the BBHE algorithm to enhance the image to be enhanced.

[0020] The beneficial effects of this invention are as follows: Addressing the problem of difficulty in enhancing the image of minute defects in VCSEL epitaxial wafers, this invention improves upon the traditional BBHE algorithm's method of segmenting grayscale histograms using grayscale mean. By fusing local, global, and structural tensor features of pixels, the defect saliency of each pixel is calculated, and a weighted histogram is constructed based on this saliency, amplifying the weight of defect information in the histogram. By finding the median energy point of this weighted histogram as the optimal segmentation threshold, it ensures that histogram segmentation is based on real defect information, thereby improving the contrast of minute defects, increasing detection accuracy, and providing a better image foundation for defect detection in epitaxial wafers. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating an image enhancement method for VCSEL epitaxial wafer detection according to the present invention; Figure 2 This is a schematic illustration of an epitaxial wafer in this invention; Figure 3 This is a schematic illustration of an epitaxial wafer image enhanced according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] This invention discloses an image enhancement method for VCSEL epitaxial wafer detection, referring to... Figure 1 This includes steps S1 to S7: S1. Acquire epitaxial wafer images and perform image preprocessing.

[0025] Since random noise is inevitably introduced during image acquisition, affecting subsequent enhancement of epitaxial wafer images, this invention acquires epitaxial wafer images, performs grayscale processing on the epitaxial wafer images to obtain grayscale images of epitaxial wafer images, and preprocesses the grayscale images using a filtering algorithm. The preprocessed grayscale images are then used as the images to be enhanced, and subsequent processing is performed on the images to be enhanced.

[0026] For example, the filtering algorithm is a bilateral filtering algorithm.

[0027] S2. Obtain the first discrimination score by measuring the grayscale changes within the neighborhood of the pixels in the image to be enhanced.

[0028] It should be noted that there is a certain visual difference between the defect location and the defect-free location in the epitaxial wafer image. Therefore, there is a certain grayscale difference between the defect location and the defect-free location in the epitaxial wafer image. In order to distinguish the defective part in the epitaxial wafer image and to better enhance the epitaxial wafer image, the present invention obtains the first discrimination degree by the grayscale change in the neighborhood of the pixel in the image to be enhanced.

[0029] Specifically, any pixel in the image to be enhanced is taken as the target pixel, and the target pixel is taken as the center. The pixel range is the neighborhood range of the target pixel, and the pixels within the neighborhood range are the neighboring pixels of the target pixel. The mean and standard deviation of the gray values ​​of the neighboring pixels of the target pixel are obtained. The mean of the gray values ​​of the neighboring pixels of the target pixel is used as the local mean, the standard deviation of the gray values ​​of the neighboring pixels of the target pixel is used as the local standard deviation, and the standard deviation of the gray values ​​of all pixels in the image to be enhanced is used as the global standard deviation. The first discrimination of the target pixel is obtained based on the local standard deviation, the global standard deviation, and the difference between the gray value of the target pixel and the local mean.

[0030] For example, It is 7.

[0031] Specifically, the first distinguishing degree satisfies the following relation: ; In the formula, For the first in the image to be enhanced The first distinguishing factor of each pixel For the first in the image to be enhanced The grayscale value of each pixel For the first in the image to be enhanced Local mean of each pixel For the first in the image to be enhanced Local standard deviation of each pixel The global standard deviation of the image to be enhanced. It is a linear normalization function.

[0032] in, Representing the The degree of dispersion of grayscale distribution within the neighborhood of a pixel. The larger the number, the more likely it is to be the first. The greater the grayscale difference between the neighboring pixels of the nth pixel, the greater the grayscale difference between the nth and nth pixels. The greater the first distinguishability of each pixel; The smaller the number, the more likely it is to be the first The smaller the grayscale difference between the neighboring pixels of the nth pixel, the better. The smaller the first distinguishability of each pixel.

[0033] Representing the Gray-level differences within the neighborhood of a single pixel The larger the number, the more likely it is to be the first. The pixel and the The greater the difference between the neighboring pixels of the nth pixel, the greater the difference between the nth and nth pixels. The easier it is to distinguish the first pixel, the more distinguishable the second pixel becomes. The greater the first distinguishability of each pixel; The smaller the number, the more likely it is to be the first The pixel and the The smaller the difference between the neighboring pixels of the nth pixel, the better. The more difficult it is to distinguish a pixel, the more difficult it is to distinguish the first pixel. The smaller the first distinguishability of each pixel.

[0034] S3. Obtain the second discrimination based on the first discrimination of the pixel and the first discrimination of the pixels in the row where the pixel is located.

[0035] It should be noted that, as Figure 2 As shown, this is an epitaxial wafer image in the present invention. The epitaxial wafer image contains certain stripe structures. Therefore, not only will the pixels at the defect location have different gray values ​​from the surrounding pixels, but there will also be certain gray value differences at the boundaries between stripe structures. In order to avoid the gray value differences at the stripe structures in the epitaxial wafer image from interfering with subsequent image enhancement, the present invention obtains a second discrimination degree based on the first discrimination degree of the pixel and the first discrimination degree of the pixels in the row where the pixel is located.

[0036] Specifically, the mean of the first discrimination of all pixels in the row containing the target pixel is obtained, and the second discrimination of the target pixel is obtained based on the difference between the first discrimination of the target pixel and the mean of the first discrimination of all pixels in the row containing the target pixel, and the first discrimination of the target pixel.

[0037] Specifically, the second discrimination degree satisfies the following relation: ; In the formula, For the first in the image to be enhanced The second distinguishing factor of each pixel For the first in the image to be enhanced The first distinguishing factor of each pixel For the first in the image to be enhanced The mean of the first discrimination of all pixels in the row containing the given pixel. To correct the coefficient, in this embodiment The value is 2, and implementers can choose a value greater than 1 as a correction coefficient based on the actual situation.

[0038] in, Representing the The pixel and the The difference in the first discriminant metric among all pixels in the same row of a given pixel is considered. Due to the horizontal stripe structure in the epitaxial image, the first discriminant metric of a pixel should generally be similar to that of the pixels in its row. To avoid the influence of the stripe structure of the epitaxial image on image enhancement, [further measures are taken]. right Perform downward correction to obtain the second discrimination; the difference between the first discrimination of a pixel and the first discrimination of the pixels in its row. The smaller, The larger, the more The difference between a pixel and its neighboring pixels is more likely to be due to the stripe structure of the epitaxial image, and should be addressed accordingly. The greater the downward correction force, the smaller the second distinction of the pixels at the stripe structure of the epitaxial film image; When it is larger, The smaller, the first The greater the difference between a pixel and its neighboring pixels, the more likely it is to be caused by defects in the epitaxial wafer. This should be addressed... The smaller the downward correction force, the less likely it is to overcorrect the first discrimination of pixels at epitaxial defects in the image, which would affect subsequent image enhancement.

[0039] It needs to be further explained that, The value range is from 0 to 1. It is greater than 1. In an exponential function, when the base is greater than 0 and less than 1, and the exponent is greater than 1, the exponent has a downward corrective effect on the base.

[0040] S4. Obtain the feature value of the pixel, and obtain the defect significance value of the pixel based on the second discrimination of the pixel and the feature value.

[0041] It should be noted that the stripe structure in the epitaxial wafer image has a single horizontal direction. Therefore, the grayscale change of the pixels at the boundary between the stripe structures is usually only in a single direction (vertically downward or vertically upward). However, the defects of the epitaxial wafer can be of any shape, and the grayscale change direction of the pixels at the defects in the image is arbitrary. The feature value obtained by the pixel structure tensor can reflect whether the grayscale change direction of the pixel is unidirectional. Therefore, this invention obtains the defect significance value of the pixel based on the second discrimination of the pixel and the feature value.

[0042] Specifically, the gradient of each pixel in the image is obtained through the Sobel operator, the structure tensor of each pixel is obtained based on the gradient of each pixel, the structure tensor of the pixel is decomposed to obtain the two feature values ​​corresponding to the pixel, and the defect significance value of the pixel is obtained based on the difference between the two feature values ​​of the pixel and the second discrimination.

[0043] It should be noted that obtaining the structure tensor through the gradient of a pixel and then obtaining the two feature values ​​of the pixel through the structure tensor are existing technologies, and will not be elaborated here.

[0044] Specifically, the significance value of the defect satisfies the following relationship: ; In the formula, For the first in the image to be enhanced The significant value of defects per pixel, For the first in the image to be enhanced The second distinguishing factor of each pixel For the first in the image to be enhanced The larger of the two feature values ​​of a pixel. For the first in the image to be enhanced The smaller of the two feature values ​​of a pixel To prevent division by zero parameters, in this embodiment... The implementers can adjust according to the actual situation. The value of .

[0045] in, It reflects the local grayscale changes of the pixel. The larger the value, the higher the value of the first element in the image to be enhanced. The more likely the grayscale change of a pixel is to be oriented in only one direction, the more likely the grayscale change of the i-th pixel is to be oriented in only one direction. The more likely a pixel is to be located at the boundary between epitaxial stripe structures in the image, the more likely the 1st pixel should be... The smaller the defect significance value of each pixel, the more effective the reduction of the influence of the stripe structure in the epitaxial wafer image on subsequent image enhancement. The smaller the value, the higher the value of the first element in the image to be enhanced. The more likely the grayscale change of a pixel is to change in more than one direction, the more likely the grayscale change of the th pixel is to change in more than one direction. The more likely a pixel is not located at the boundary between epitaxial stripe structures in the image, the more likely the 1st pixel should be made to be... The larger the defect significance value of the first pixel, the better the improvement of the first pixel. The influence of each pixel on the subsequent image enhancement process is analyzed to achieve better enhancement of the epitaxial wafer image. For ease of calculation, this is done by... Divide by The way to Normalization was performed.

[0046] Second Discrimination This reflects the grayscale variation of a pixel within its row. The larger the value, the higher the value of the first element in the image to be enhanced. The more likely a pixel is to be located at a grayscale change point outside the boundary between stripe structures in the epitaxial image, the more likely it is to be a pixel at that point. The higher the probability that the nth pixel is located at an epitaxial wafer defect, the higher the probability that the nth pixel in the image to be enhanced will be. The larger the defect value of a pixel, the greater its significance. The smaller the value, the higher the value of the image to be enhanced. The more likely a pixel is to be at a defect-free location, such as the boundary between stripe structures in the epitaxial image, the better the image to be enhanced will be. The smaller the defect significance value of each pixel, the better.

[0047] S5. Construct a weighted histogram based on the defect significance values ​​of the pixels.

[0048] It should be noted that in existing BBHE algorithms, the histogram statistics are applied equally to all pixels. Each pixel is represented by a bar with a cumulative gray level value of 1. For example, if there are 10 pixels with a gray level value of 100 in an image, the bar corresponding to gray level value 100 in the histogram will be 10. This means that the vast majority of non-defective pixels in the image dominate the shape of the histogram. Therefore, the existing BBHE algorithm has limited image enhancement effects. To better enhance epitaxial wafer images, this invention constructs a weighted histogram based on the defect significance value of each pixel.

[0049] Specifically, the weighted value of each pixel is obtained based on the defect significance value of each pixel in the image to be enhanced. The weighted values ​​of all pixels corresponding to each gray level in the image to be enhanced are accumulated, and the accumulated results are arranged in ascending order of gray level to obtain a weighted histogram.

[0050] For example, there are 10 pixels with a gray value of 100 in the image to be enhanced. The weighted values ​​of these 10 pixels are (1.2, 1.3, 1.2, 1.4, 1.1, 1.5, 1.2, 1.3, 1.2, 1.4). Then the height of the bar at gray level 100 in the weighted histogram is 1.2+1.3+1.2+1.4+1.1+1.5+1.2+1.3+1.2+1.4=12.8.

[0051] Specifically, the column height of each gray level in the weighted histogram satisfies the following relationship: ; In the formula, For the weighted histogram, the first A column height of one gray level, The grayscale value in the image to be enhanced is The The significant value of defects per pixel, The grayscale value in the image to be enhanced is The number of pixels, In this embodiment, the weighting factor is the histogram weighting factor. The weighting factor is set to 2, and the implementers can adjust the size of the weighting factor according to the actual situation.

[0052] Where 1 represents the basic weighting value of a pixel. Since the BBHE algorithm calculates the histogram by accumulating one unit of bar height for each pixel, 1 is used as the basic weighting value. For the first in the image to be enhanced The corresponding gray level is the th The weighted value of each pixel, The larger the value, the higher the value of the image. The corresponding gray level is the th The more likely a pixel is to correspond to a defect in the epitaxial wafer, the larger the weighting value of the pixel should be, so that the pixel corresponding to the defect in the epitaxial wafer has a greater influence on the weighted histogram, making it easier to enhance the image features of the defect in the epitaxial wafer when performing image enhancement. The smaller the value, the higher the value in the image. The corresponding gray level is the th The more likely a pixel is to correspond to a defect-free area of ​​the epitaxial wafer, the smaller its weighting value should be. This would minimize the influence of defect-free pixels on the weighted histogram of the epitaxial wafer image, thereby reducing the impact of defect-free pixels on the enhancement of the epitaxial wafer image.

[0053] S6. Obtain the segmentation threshold based on the weighted histogram.

[0054] It should be noted that current techniques for the BBHE algorithm only use the mean gray level of the image as the segmentation point, and then calculate the gray level histogram. Figure 1 The image enhancement is divided into two parts, which lacks targeted enhancement of the epitaxial wafer image. In order to better enhance the epitaxial wafer image, this invention obtains the segmentation threshold based on the weighted histogram.

[0055] Specifically, the sum of the heights of all gray levels in the weighted histogram is obtained, and half of the sum of the heights of all gray levels is taken as the median of the significant energy. Any gray level in the weighted histogram is taken as the target gray level, and the cumulative sum of the heights of all gray levels corresponding to the target gray level and the first gray level (including the target gray level and the first gray level) is obtained. This cumulative sum is taken as the significant energy value of the target gray level, and the difference between the significant energy value of the target gray level and the median of the significant energy is taken as the energy difference of the target gray level. The value corresponding to the gray level with the smallest energy difference is taken as the segmentation threshold of the weighted histogram.

[0056] For example, the column height corresponding to gray level 0 is 10, the column height corresponding to gray level 1 is 15, the column height corresponding to gray level 2 is 20, and the median significant energy is 100. The difference between the sum of the column heights corresponding to gray level 0 and gray level 1 and the median significant energy is obtained to get the energy difference of gray level 1. This value is... The difference between the sum of the column heights corresponding to gray level 0, gray level 1, and gray level 2, and the median significant energy, is obtained to determine the energy difference for gray level 2. This value is... This process is repeated to obtain the energy difference for each gray level, and the value corresponding to the gray level with the smallest energy difference is used as the segmentation threshold. If the energy difference for gray level 150 is the smallest, then the segmentation threshold is 150.

[0057] Specifically, the energy difference satisfies the following relationship: ; In the formula, For the weighted histogram, the first The energy difference of each gray level For the weighted histogram, the first A column height of one gray level, It represents a significant energy median.

[0058] Among them, the median of significant energy This represents half the sum of the heights of all bars in the weighted histogram. For the first The sum of the column heights corresponding to the given gray level and all preceding gray levels. The smaller the value, the better. The value corresponding to each gray level is the segmentation threshold. The more likely the weighted histogram is to be divided into two parts with the same sum of bar heights, the more energy the defect area and the background area will be balanced during the image enhancement process. At this time, no matter which part the gray level corresponding to the epitaxial wafer defect corresponds to, the defect part of the epitaxial wafer can be effectively enhanced, and the targeting of the enhancement effect can be improved. The larger the value, the more likely it is to be the first The value corresponding to each gray level is less likely to divide the weighted histogram into two parts with equal sum of bar heights. In this case, if the epitaxial wafer defect corresponds to the part with a smaller sum of bar heights, the enhancement effect on the epitaxial wafer defect will be limited, making the enhancement effect worse. Therefore, the energy difference is calculated, and the value corresponding to the gray level with the smallest energy difference is used as the segmentation threshold.

[0059] S7. Perform image enhancement on the image to be enhanced by using a segmentation threshold, and then perform epitaxial wafer defect detection on the enhanced image.

[0060] Specifically, the grayscale histogram of the image to be enhanced is obtained. The grayscale histogram of the image to be enhanced is segmented using the BBHE algorithm with a segmentation threshold. The existing technology of the BBHE algorithm is then used to enhance the image. The enhanced epitaxial wafer image is used as the input image for epitaxial wafer defect detection.

[0061] Indicative Figure 3 This is an enhanced epitaxial wafer image in this invention. The enhanced epitaxial wafer image in this invention has higher contrast, and the defects of the epitaxial wafer can be identified more accurately by performing epitaxial wafer defect detection through the enhanced epitaxial wafer image.

Claims

1. An image enhancement method for VCSEL epitaxial wafer inspection, characterized in that, include: Taking any pixel in the image to be enhanced as the target pixel, the first discrimination of the target pixel is obtained based on the local standard deviation, the global standard deviation, and the difference between the gray value of the target pixel and the local mean. The local standard deviation is the standard deviation of the gray values ​​of the neighboring pixels of the target pixel; the global standard deviation is the standard deviation of the gray values ​​of all pixels in the image to be enhanced; and the local mean is the mean of the gray values ​​of the neighboring pixels of the target pixel. The second discrimination of the target pixel is obtained based on the difference between the first discrimination of the target pixel and the mean of the first discrimination of the pixels in the row where the target pixel is located, and the first discrimination of the target pixel; the defect significance value of the target pixel is obtained based on the difference between the two feature values ​​obtained from the structure tensor of the target pixel, and the second discrimination. The weighted values ​​of all pixels corresponding to each gray level in the image to be enhanced are accumulated, and a weighted histogram is obtained based on the accumulation result; the weighted values ​​are positively correlated with the defect significance value; the cumulative sum of the column heights corresponding to all gray levels between each gray level and the first gray level in the weighted histogram is obtained, and the segmentation threshold is obtained based on the cumulative sum of the column heights corresponding to the gray levels; Image enhancement is performed on the image to be enhanced using the BBHE algorithm based on the segmentation threshold.

2. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The first discrimination satisfies the following relation: ； In the formula, For the first image to be enhanced The first distinguishing factor of each pixel For the first image to be enhanced The grayscale value of each pixel For the first image to be enhanced Local mean of each pixel For the first image to be enhanced Local standard deviation of each pixel The global standard deviation of the image to be enhanced. It is a linear normalization function.

3. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The second discrimination index satisfies the following relation: ； In the formula, For the first image to be enhanced The second distinguishing factor of each pixel For the first image to be enhanced The first distinguishing factor of each pixel For the first image to be enhanced The mean of the first discrimination of all pixels in the row containing the given pixel. This is a correction factor.

4. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The significant value of the defect satisfies the following relationship: ； In the formula, For the first image to be enhanced The significant value of defects per pixel, For the first image to be enhanced The second distinguishing factor of each pixel For the first image to be enhanced The larger of the two feature values ​​of a pixel. For the first image to be enhanced The smaller of the two feature values ​​of a pixel To prevent division by zero parameters.

5. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The step of obtaining the weighted histogram based on the accumulated results includes: sorting the accumulated weighted values ​​of all pixels corresponding to each gray level in ascending order of gray level to obtain the weighted histogram.

6. An image enhancement method for VCSEL epitaxial wafer detection according to claim 1 or 5, characterized in that, The method for obtaining the weighted value includes: pre-setting a histogram weighting factor, and adding 1 to the result of multiplying the defect significance value by the histogram weighting factor to obtain the weighted value.

7. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The column height of each gray level in the weighted histogram satisfies the following relationship: ； In the formula, For the weighted histogram, the first A column height of one gray level, The grayscale value in the image to be enhanced is The The significant value of defects per pixel, The grayscale value in the image to be enhanced is The number of pixels, This is the weighting factor for the histogram.

8. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The step of obtaining the segmentation threshold based on the cumulative sum of the column heights corresponding to gray levels includes: obtaining the sum of the column heights of all gray levels in the weighted histogram, and taking half of the sum of the column heights of all gray levels as the median of significant energy; obtaining the cumulative sum of the column heights corresponding to all gray levels between each gray level and the first gray level in the weighted histogram, taking the difference between the cumulative sum and the median of significant energy as the energy difference, and taking the value corresponding to the gray level with the smallest energy difference as the segmentation threshold.

9. The image enhancement method for VCSEL epitaxial wafer detection according to claim 8, characterized in that, The energy difference satisfies the following relationship: ； In the formula, For the weighted histogram, the first The energy difference of each gray level For the weighted histogram, the first A column height of one gray level, It represents a significant energy median.

10. The image enhancement method for VCSEL epitaxial wafer detection according to claim 1, characterized in that, The method of using the BBHE algorithm to enhance the image to be enhanced based on the segmentation threshold includes: segmenting the gray-level histogram of the image to be enhanced using the segmentation threshold, dividing the gray-level histogram into two parts, and then using the existing technology of the BBHE algorithm to enhance the image to be enhanced.

Images