WO2021027163A1 - Reticulate pattern-containing image recognition method and apparatus, and terminal device and medium - Google Patents

Abstract

A reticulate pattern-containing image recognition method and apparatus, and a terminal device and a medium, suitable for the technical field of data processing. The method comprises: inputting an image to be processed into a pre-trained reticulate pattern removal model to obtain a reticulate pattern-free image to be processed (S101); performing gray value difference calculation on the image to be processed and the reticulate pattern-free image to be processed, and performing reticulate pattern reconstruction on the basis of the calculated gray value difference and a preset difference threshold to obtain a corresponding reticulate pattern graph (S102); calculating the number of pixel points included in the reticulate pattern graph, and calculating the proportion value of the obtained number of pixel points in the total number of pixel points of the image to be processed; performing graph matching on the reticulate pattern graph on the basis of a preset reticulate pattern graph library (S103); and if the number of reticulate pattern pixel points is greater than a preset number threshold, the proportion value is greater than a preset proportion threshold, and the matching of the reticulate pattern graph succeeds, determining that the image to be processed is a reticulate pattern-containing image (S104). The method ensures the accuracy and reliability of reticulate pattern-containing image recognition.

Description

Recognition method, device, terminal equipment and medium for reticulated image

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on August 9, 2019, the application number is 201910736543.2, and the invention title is "A method, device and terminal device for reticulated image recognition", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application belongs to the field of data processing technology, and in particular relates to a method, device, terminal equipment and medium for recognizing a textured image.

Background technique

Some of the ID photos of the public security system have textures. When performing face recognition and other processing, the textures will greatly affect the accuracy of recognition. Therefore, ID photos with textures need to be reticulated before they can be used. Before removing the texture, you need to determine whether the ID photo has texture. Although there are some methods for recognizing moire images in the prior art, the recognition accuracy is not ideal. Therefore, a method that can accurately identify whether an image has moire is needed.

technical problem

In view of this, the embodiments of the present application provide a method, device, terminal equipment, and medium for recognizing a textured image, which can solve the problem of low accuracy of recognizing a textured image.

Technical solutions

The first aspect of the embodiments of the present application provides a method for recognizing a textured image, including:

Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the mesh in the image;

Perform gray value difference calculation on the to-be-processed image and the to-be-processed image without texture, and perform texture reconstruction based on the calculated gray value difference and a preset difference threshold to obtain the corresponding texture Graphics

Counting the number of pixels contained in the texture pattern, and calculating the ratio of the number of pixels to the total number of pixels in the image to be processed; performing graphic matching on the texture pattern based on a preset texture pattern library;

If the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.

A second aspect of the embodiments of the present application provides a texture image recognition device, including:

The de-texturing module is used to input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is based on the reticulated image sample and the non-reticulated image sample in advance. The trained model is used to remove the moire in the image;

The reticulation reconstruction module is used to calculate the gray value difference of the image to be processed and the non-reticulated image to be processed, and perform reticulation based on the calculated gray value difference and a preset difference threshold Reconstruct to get the corresponding net pattern;

The feature processing module is used to count the number of pixels contained in the texture pattern, and calculate the ratio of the number of pixels to the total number of pixels in the image to be processed; and compare the texture pattern based on a preset texture pattern library Perform graphic matching;

The mesh discrimination module is configured to determine that the image to be processed is a mesh image if the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully.

A third aspect of the embodiments of the present application provides a terminal device, including a memory and a processor. The memory stores computer-readable instructions that can run on the processor. The processor executes the computer The following steps are implemented when reading instructions:

Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the mesh in the image;

Perform gray value difference calculation on the to-be-processed image and the to-be-processed image without texture, and perform texture reconstruction based on the calculated gray value difference and a preset difference threshold to obtain the corresponding texture Graphics

Counting the number of pixels contained in the texture pattern, and calculating the ratio of the number of pixels to the total number of pixels in the image to be processed; performing graphic matching on the texture pattern based on a preset texture pattern library;

If the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.

A fourth aspect of the embodiments of the present application provides a computer-readable storage medium that stores computer-readable instructions, wherein the computer-readable instructions are implemented when executed by at least one processor The following steps:

Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the mesh in the image;

Perform gray value difference calculation on the to-be-processed image and the to-be-processed image without texture, and perform texture reconstruction based on the calculated gray value difference and a preset difference threshold to obtain the corresponding texture Graphics

Counting the number of pixels contained in the texture pattern, and calculating the ratio of the number of pixels to the total number of pixels in the image to be processed; performing graphic matching on the texture pattern based on a preset texture pattern library;

If the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.

Beneficial effect

The pre-trained reticulation removal model is used to remove reticulation of the image to be processed to obtain a non-reticulated image to be processed, and then the gray value difference of the to-be-processed image before and after removal is compared to determine the reticulation The differences between the images to be processed before and after are drawn as corresponding mesh patterns (that is, the difference is assumed to be mesh). In actual situations, the mesh pattern of the ID photo has the following characteristics: 1. The size of the ID photo is relatively fixed , So the number of pixels occupied by the texture is relatively stable. 2. In the ID photo, the mesh generally covers most or the entire area of the image, so the proportion of pixels occupied by the mesh is relatively stable and high. 3. The types of texture patterns placed on ID photos are limited and known. Based on the above-mentioned practical characteristics, the embodiments of this application will further verify the number of pixels contained in the drawn texture pattern and the proportion of the total pixel points in the image to be processed, and match the texture pattern to achieve Check the multi-dimensional all aspects of the mesh, when all are satisfied, it means that the difference between the image to be processed before and after removing the mesh is the content of the mesh, so the image to be processed can be determined to be a mesh image, ensuring the mesh image The recognition is accurate and reliable.

Description of the drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only of the present application. For some embodiments, for those of ordinary skill in the art, other drawings can be obtained from these drawings without creative labor.

FIG. 1 is a schematic diagram of the implementation process of the moire image recognition method provided by Embodiment 1 of the present application;

FIG. 2 is a schematic diagram of the implementation process of the moire image recognition method provided in Embodiment 2 of the present application;

FIG. 3 is a schematic diagram of the implementation process of the moire image recognition method provided by Embodiment 3 of the present application;

FIG. 4 is a schematic diagram of the structure of the texture image recognition device provided by the fourth embodiment of the present application;

FIG. 5 is a schematic diagram of a terminal device provided in Embodiment 5 of the present application.

Embodiments of the invention

In the following description, for the purpose of illustration rather than limitation, specific details such as a specific system structure and technology are proposed for a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from obstructing the description of this application.

In order to illustrate the technical solutions described in the present application, specific embodiments are used for description below.

Since an image with mesh will greatly affect the accuracy of face recognition and other processing on the image, a method that can identify whether an image has mesh is needed to provide a basis for mesh removal and face recognition in ID photos.

In order to realize the reticulation of the image, in the embodiment of the application, the pre-trained reticulation removal model is used to perform reticulation removal on the image to be processed to obtain a non-reticulated image to be processed, and then the to-be-processed image before and after the removal Carry out the gray value difference comparison, so as to determine the difference part of the image to be processed before and after the netting, and draw it as the corresponding net pattern (that is, assuming the difference is the net pattern), because the ID photo has the net pattern in the actual situation The following features: 1. The size of the ID photo is relatively fixed, so the number of pixels occupied by the mesh is relatively stable. 2. In the ID photo, the mesh generally covers most or the entire area of the image, so the proportion of pixels occupied by the mesh is relatively stable and high. 3. The types of texture patterns placed on ID photos are limited and known. Based on the above-mentioned practical characteristics, the embodiments of this application will further verify the number of pixels contained in the drawn texture pattern and the proportion of the total pixel points in the image to be processed, and match the texture pattern to achieve Check the multi-dimensional all aspects of the mesh, when all are satisfied, it means that the difference between the image to be processed before and after removing the mesh is the content of the mesh, so the image to be processed can be determined to be a mesh image, ensuring the mesh image The recognition is accurate and reliable.

FIG. 1 shows the implementation flowchart of the moire image recognition method provided by Embodiment 1 of the present application, and the details are as follows:

S101: Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the moiré in the image.

In the embodiment of this application, the mesh removal model is used for the mesh in the image, which is pre-trained and constructed by the technicians. After the mesh removal model is processed, it will be obtained regardless of whether the original image to be processed contains mesh or not. The corresponding non-textured image. Among them, the training and construction method of the reticulation removal model is not limited here, and can be designed by the technicians themselves, or the training and construction can also be carried out by referring to Embodiments 2 to 6 of this application.

S102: Perform gray value difference calculation on the image to be processed and the non-textured image to be processed, and perform reticulation reconstruction based on the calculated gray value difference and a preset difference threshold to obtain a corresponding reticulated pattern.

In order to identify whether the image to be processed contains mesh, the embodiment of the present application will first assume that the image to be processed contains mesh, and will directly calculate the gray value difference of the image before and after the mesh is removed. Each pixel corresponding to the image is grayed out and the gray value difference is performed to determine the difference between the image to be processed before and after the texture is removed. At the same time, even if it is not a texture pixel, it is removed through the texture There may also be changes in the gray value before and after the model is processed. Therefore, in this embodiment of the application, a difference threshold is set in advance to select mesh pixels, and only the pixels whose gray value difference is greater than the difference threshold are set Recognized as mesh pixels. Among them, the specific size of the difference threshold can be set by a technician according to actual application requirements.

After the reticulated pixels are selected, the graphics composed of all the reticulated pixels are extracted as the corresponding reticulated graphics, so as to realize the reconstruction of the reticulated in the original image to be processed.

S103: Count the number of pixels included in the texture pattern, and calculate the ratio of the number of pixels to the total number of pixels of the image to be processed. Perform graphic matching on the reticulated graphic based on the preset reticulated graphic library.

S104: If the number of mesh pixels is greater than the preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.

Due to the actual situation, ID photo reticulation has the following characteristics:

1. The size of the ID photo is relatively fixed, so the number of pixels occupied by the mesh is relatively stable.

2. In the ID photo, the mesh generally covers most or the entire area of the image, so the proportion of pixels occupied by the mesh is relatively stable and high.

3. The types of texture patterns placed on ID photos are limited and known.

It is precisely in combination with the above three characteristics of the reticulation of the ID photo that, in this embodiment of the application, the number of reticulated pixels contained in the ID photo that actually contains the reticulation and the proportion of pixels in the total ID photo image will be counted in advance, and According to the statistical results, the corresponding number threshold and proportion threshold are set. At the same time, based on the known types of reticulated patterns of ID photos, a corresponding reticulated pattern library is constructed. In the actual processing, the reticulated graphics obtained by the drawing were counted and the ratio of the total pixel points of the image to be processed was calculated, and compared with the corresponding number threshold and ratio threshold, and at the same time according to the drawn The reticulated graphics matches the reticulated graphic library. If the processing result is that the number of reticulated pixels is large enough and the proportion is large enough, and there are similar graphics in the reticulated graphic library, it indicates the difference before and after removal of the reticulated graphic library. The content also satisfies the three characteristics of the mesh of the above-mentioned ID photo, that is, the assumption that the content of the difference is a mesh is established. At this time, the image to be processed is directly determined to be a mesh image to complete the recognition of the mesh image.

Among them, if the three conditions are not met, it means that the detection result cannot determine whether the image to be processed contains mesh. The reason for this result is not only that the image to be processed does not contain mesh, but also There may be problems such as data errors during processing, or poor image quality to be processed. Therefore, as an optional embodiment of this application, if the three conditions are not met, the embodiment of this application will return to S101 Reprocess the image to be processed, and count the total number of processing of the image to be processed at the same time. If the processing result is within the preset maximum total number of times, the processing result is that one of the three conditions is not met, then loop back to S101 until the processing times When the maximum total number of times is reached, it is directly determined that the image to be processed is an image without mesh. As another optional embodiment of the present application, if the three conditions are not met, it is also possible to directly determine that the image to be processed is an image without mesh. The specific method can be selected and set by the technicians according to actual needs, and is not limited here.

In the embodiment of the application, the pre-trained reticulation removal model is used to perform reticulation removal on the image to be processed to obtain a non-reticulated image to be processed, and then the gray value difference ratio of the to-be-processed image before and after the removal Yes, so as to determine the difference part of the image to be processed before and after the netting, and draw it as the corresponding net pattern (that is, assuming that the difference part is the net pattern), and then based on the three characteristics of the ID photo, further analyze the drawn net The number of pixels contained in the texture pattern and the proportion of the total pixel points in the image to be processed are verified, and the texture pattern is matched to achieve a multi-dimensional full-scale check of the texture. When all are satisfied, the texture is removed. The difference between the images to be processed before and after is the mesh content, so the image to be processed can be determined to be a mesh image, which ensures the accuracy and reliability of the mesh image recognition.

As a specific implementation of the training and construction of the mesh removal model in the first embodiment of the present application, as shown in FIG. 2, the second embodiment of the present application includes:

S201. Obtain multiple pairs of textured image samples and non-reticulated image samples, where each pair of image samples includes a textured image sample and a non-reticulated image sample, and the textured image sample in each pair of image samples There is only a difference in texture between the sample and the sample without texture.

In the embodiment of this application, the image samples for model training are all in pairs, and the mesh image samples and non- mesh image samples in each pair of image samples are identical except for the mesh. In order to obtain For multiple pairs of image samples with only differences in texture, the methods used include, but are not limited to, obtaining the required number of non-textured ID photos, and then adding corresponding textures to these non-textured ID photos, or technical personnel Other methods can also be used, which are not limited here.

S202, construct a texture removal generator G(x), a texture addition generator F(x), set G(x) through the discriminating network Dg(x) to obtain the probability that the image belongs to a non-textured image as Dg(G(x) )), F(x) obtains the probability that the image belongs to the texture image through the discriminant network Df(x) as Df(F(x)).

In the embodiment of this application, the initial overall model will be constructed first, and the overall model includes the initial texture removal generator G(x), texture addition generator F(x), non-texture discrimination network Dg(x), and The reticulated discriminant network Df(x) is used for subsequent iterative training. The initial model construction rules include but are not limited to the following: The model frame structure, how many layers it contains, the attributes of each layer, etc., are set by the technicians, and random Generate model parameters. The initial recognition rates of G(x) and F(x) are generally low. Therefore, in the embodiments of the present application, subsequent iterative update training will be used to improve the model recognition rate.

S203: Use G(x) and F(x) to process the texture image sample a and the non-texture image sample b respectively to obtain the corresponding processed image, and based on a, b, processed image, Dg(G( a)) and Df(F(b)) calculate the first loss value corresponding to G(x) and F(x), and calculate Dg(x) and Df based on Dg(G(a) and Df(F(b)) (x) The corresponding second loss value and third loss value.

In order to realize the effectiveness evaluation of the texture removal generator G(x), the texture addition generator F(x), the non-texture discrimination network Dg(x), and the texture discrimination network Df(x), embodiments of the present application We will use G(x) and F(x) to process the textured image sample a and the non-reticulated image sample b respectively to obtain the corresponding processed image. Suppose G(x) is used for the textured image sample a Process to get image a', use F(x) to process a'to get image a", use F(x) to process non-textured image sample b to get image b', use F(x) to process b' The processed image b" is obtained. The processed image in the embodiment of the present application includes but is not limited to one or more of a', a", b', and b". In the embodiment of this application, firstly, G(x) and F(x) are used to process a and b respectively. Since a and b only have the difference in texture, the theoretically obtained processed image should also be only There is a difference in reticulation. In the same way, we can see that in theory a=a"=b' and b=b"=a'. Based on this theoretical equation, the embodiment of this application will further calculate the generator loss of the two functions. The first loss value of the function, and the second and third loss values corresponding to Dg(x) and Df(x). The selection/design of the specific loss function is not limited here, and can be selected by the technicians according to their needs /Design, you can also refer to the second and third embodiments of this application for processing.

S204: Calculate the degree of image difference between a, b and the processed image.

Although theoretically a=a"=b' and b=b"=a', the processing effect of untrained G(x) and F(x) will not be very good in actual situations, so the actual a , A" and b', and between b, b" and a', there must be a certain difference, and this difference directly reflects the training effect of G(x) and F(x), so the embodiment of the application will The method of calculating the image difference between a, b and the processed images is used as a quantification method of one dimension of the training effect of G(x) and F(x). Among them, the specific calculation method of the image difference degree is not limited here, including but not limited to calculating the Euclidean distance between a, a" and b', and the Euclidean distance between b, b" and a' It can also be designed by the technicians according to their needs. At the same time, it should be noted that depending on the final calculation method, there will be differences in the images used to the specific processing. For example, only a, b, a can be used. 'And b'for calculation, you can also use a, b, a', b', a" and b" for calculation at the same time, and the specific final calculation method is determined.

In order to realize the iterative training of G(x), F(x), Dg(x) and Df(x) to achieve the expected training effect, the embodiment of this application will preset one or more loss value thresholds, and set them in advance A good difference threshold is used to judge the validity of the three loss values and the difference degree of the image. The loss threshold is used to measure the expected training effects of G(x), F(x), Dg(x) and Df(x). The number of them is set by the technicians according to actual needs. When the generator and the discriminant network When the expected effects of training are different, an independent loss threshold can be set for each loss value. Similarly, the same loss threshold can also be set uniformly. At the same time, the specific values of the loss threshold and the difference threshold can also have technical personnel Set according to actual needs, and the larger the loss threshold and the difference threshold, the lower the requirements for the expected effect of training the generator and the discriminant network.

S205: If the second loss value and/or the third loss value are substantially equal to the corresponding preset loss value threshold value, iteratively update Dg(x) and Df(x).

S206: If the first loss value is greater than the corresponding preset loss value threshold and/or the image difference degree is greater than the preset difference degree threshold, iteratively update G(x) and F(x).

When there is a value that does not meet the loss threshold requirement in the second loss value and the third loss value, it means that the discriminating effect of Dg(x) and Df(x) has reached the expected effect, so it will return to iterative update Dg(x ) And Df(x). Similarly, when the first loss value is too large and does not meet the loss value threshold requirement, it means that G(x) and F(x) have not achieved the expected effect, and iteratively update G(x) and F(x) at this time.

S207: If the first loss value, the second loss value, and the third loss value are all less than the corresponding preset loss value threshold, and the image difference degree is less than the preset difference degree threshold, complete the image removal generator G(x) The model is trained to get the reticulation removal model.

Since G(x), F(x), Dg(x), and Df(x) are mutually opposed and dependent on each other, in this embodiment of the application even if the second loss value and the third loss value both meet the requirements , Or the first loss value meets the requirements, it cannot be directly determined that the training of Dg(x) and Df(x) is completed, or the training of G(x) and F(x) is completed. Therefore, in the embodiment of this application, only When the first loss value, the second loss value, the third loss value and the image difference degree meet the requirements at the same time, it will be judged that the training of G(x), F(x), Dg(x) and Df(x) is completed. At this time, the final usable texture removal generator G(x) can be obtained, that is, the texture removal model in the first embodiment of the present application.

It should be noted that although our ultimate goal is to train and construct G(x) that can be reticulated to obtain the reticulated removal model of the first embodiment of this application, in this embodiment of the present application, G(x ) Training and training effect judgment is not only closely related to the relative function of F(x), but also closely related to the accuracy of the network Dg(x) and Df(x) to determine whether there is reticulation. Only in G(x), Only when F(x), Dg(x) and Df(x) have completed the training to achieve the expected effect, can it be explained that the final G(x) is accurate and effective. Therefore, the embodiments of the present application compare G(x) and F(x) ) While iteratively updating, Dg(x) and Df(x) will also be iteratively updated, although the steps of Dg(x) and Df(x) update seem to be independent (the judgment of whether to update depends only on the second loss value In the case of the third loss value and the third loss value, there is no reference to the first loss value and the image difference degree), but in fact the second loss value and the third loss value also depend on the G(x) and F(x) pairs of images updated in real time The sample processing effect, so in fact, the update of Dg(x) and Df(x) and G(x) and F(x) are inseparable from each other and cannot be simply regarded as two independent iterative updates. Steps.

In the embodiment of this application, by constructing two opposing generators and two opposing discriminant networks, the paired image samples are processed based on the opposing generators and discriminant networks, and the loss value and image are calculated based on the processing results. The calculation of the difference degree is to realize the quantification of the training effect of the generator and the discriminant network. Finally, the generator and the discriminant network are respectively updated iteratively according to whether the loss value and the image difference degree meet the expected effect, until they both meet the expected effect. The effective training of the reticulation removal model is realized.

As a specific implementation manner of calculating the first loss value in the second embodiment of the present application, it includes:

Use G(x) to process the image a'obtained by processing the textured image sample a, use F(x) to process the image a" obtained from a', and use F(x) to process the non-textured image sample b to obtain The first loss value is calculated based on the formula (1) for the image b′ obtained by processing b′ by F(x):

Lg=-(log ₁₀ (Dg(G(a)))-log ₁₀ (Df(F(b)))+Lcyc,

Lcyc=L1Loss(a″,a)×lambda_a+L1Loss(b″,b)×lambda_b+

L1 Loss(a,b')×lambda_c+L1 Loss(b,a')×lambda_d (1)

Among them, Lg is the first loss value, L1 Loss(x, y) represents the Euclidean distance of the two images, and lambda_a, lambda_b, lambda_c, and lambda_d represent preset weights.

Since theoretically a=a"=b', and b=b"=a', in the embodiments of this application, a" and a, b" and b, a and b', and b and a'will be compared respectively Calculate the corresponding Euclidean distance to obtain the quantified value of the difference degree of the corresponding four dimensions. At the same time, in theory, the more the image sample generator processes, the greater the probability of deviation, and the more difficult it is to match the original image sample. Therefore, in the embodiment of the present application, corresponding weights are preset for the degree of difference of different dimensions to balance the difference degree values obtained from different matching difficulty situations, and to ensure the effectiveness of the first loss value calculation to the greatest extent Sex. Among them, the specific values of lambda_a, lambda_b, lambda_c, and lambda_d can be set by a technician after measuring the matching difficulty of each dimension. Preferably, the values of lambda_a and lambda_b are greater than lambda_c and lambda_d.

As a specific implementation manner of calculating the second loss value and the third loss value in the second embodiment of the present application, it includes:

Calculate the second loss value Ldg and the third loss value Ldf based on formula (2) and formula (3):

Ldg=-log ₁₀ (Dg(G(a))-0.5)+log ₁₀ (1.5-Dg(G(a))) (2)

Ldf=-log ₁₀ (Df(F(b))-0.5)+log ₁₀ (1.5-Df(F(b))) (3)

As a specific implementation of calculating the degree of image difference in the second embodiment of the present application, as shown in FIG. 3, the third embodiment of the present application includes:

S301. Perform gray value difference calculations on a and b, a and a', and b and b'respectively, and perform mesh extraction based on the obtained gray value difference and a preset difference threshold to obtain the corresponding first Reticulated image, second reticulated image and third reticulated image, where a'is the image obtained by processing the reticulated image sample a using G(x), and b'is the non-reticulated image using F(x) The image obtained by processing the image sample b.

S302. Calculate the image distance between the first mesh image and the second mesh image, and the image distance between the first mesh image and the third mesh image, and calculate the difference between the two image distances. Get the image difference degree.

Since in theory a=a"=b' and b=b"=a', the first mesh image is a standard mesh image, and the second mesh image is the difference part image before and after G(x) processing. The three-screen image is the different part of the image before and after the F(x) processing. Therefore, the gray value difference operation is performed on a and a'and b and b'respectively to realize the actual processing of G(x) and F(x) The effect is extracted, and then the image distances between the second and third mesh images and the first mesh image are respectively calculated. Several realizing the quantitative evaluation of the actual processing effects of G(x) and F(x), and finally By calculating the difference between the two image distances, the degree of image difference required in the second embodiment of the present application can be obtained. Wherein, the image distance is the reciprocal of the similarity of the image, and the specific image distance calculation method is not limited here, and can be set by the technicians themselves, or refer to other embodiments of this application.

As a specific implementation of calculating the image distance in the third embodiment of the present application, it includes:

Calculate the image distance L based on formula (4):

Wherein, n is the total number of pixels of the first mesh image or the second mesh image, and x _i and y _i are the _ith of the first mesh image or the second mesh image, respectively The pixel value of each pixel.

In the embodiment of the present application, the pixel points of the texture image are compared one by one to find the pixel value difference, and then the difference is taken as the reciprocal to obtain the required image distance. In the same way, the image distance between the first mesh image and the third mesh image can also be calculated according to formula (4), which will not be repeated here.

Corresponding to the method of the above embodiment, FIG. 4 shows a structural block diagram of a moire image recognition device provided in an embodiment of the present application. For ease of description, only the parts related to the embodiment of the present application are shown. The texture image recognition device illustrated in FIG. 4 may be the execution subject of the texture image recognition method provided in the first embodiment.

4, the texture image recognition device includes:

The de-interlacing module 41 is used to input the image to be processed into a pre-trained de-interlacing model to obtain a de-interlacing image to be processed, and the de-interlacing model is based on the deinterlacing image sample and the non-interlacing image sample The model obtained by training is used to remove moire in the image.

The mesh reconstruction module 42 is used to calculate the gray value difference between the image to be processed and the image without mesh to be processed, and perform mesh based on the calculated gray value difference and a preset difference threshold. Reconstruction of the texture to obtain the corresponding texture pattern.

The feature processing module 43 is configured to count the number of pixels included in the texture pattern, and calculate the ratio of the number of pixels to the total number of pixels of the image to be processed. Graphic matching is performed on the mesh pattern based on a preset mesh pattern library.

The mesh discrimination module 44 is configured to determine that the image to be processed is a mesh image if the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully.

Further, the texture image recognition device further includes:

The sample acquisition module is used to obtain multiple pairs of textured image samples and non-reticulated image samples, where each pair of image samples includes a textured image sample and a non-reticulated image sample, and each pair of image samples There is only a texture difference between the texture image sample and the non-texture image sample in.

The generator building module is used to construct the reticulation removal generator G(x), and the reticulation addition generator F(x). Let G(x) obtain the probability that the image belongs to the non-recreation image through the discrimination network Dg(x) as Dg(G(x)), F(x) through the discriminant network Df(x), the probability that the image belongs to the texture image is Df(F(x)).

Loss value calculation module, used to use G(x) and F(x) to process the texture image sample a and the non-texture image sample b respectively to obtain the corresponding processed image, and based on a, b, and the processing After image, Dg(G(a)) and Df(F(b)) calculate the first loss value corresponding to G(x) and F(x), based on Dg(G(a) and Df(F(b)) Calculate the second loss value and the third loss value corresponding to Dg(x) and Df(x).

The difference calculation module is used to calculate the image difference degree between a, b and the processed images.

The iterative update module is configured to iteratively update Dg(x) and Df(x) if the second loss value and/or the third loss value are substantially equal to the corresponding preset loss value threshold. If the first loss value is greater than the corresponding preset loss value threshold and/or the image difference degree is greater than the preset difference degree threshold, G(x) and F(x) are updated iteratively.

A model output module, configured to: if the first loss value, the second loss value, and the third loss value are all less than a corresponding preset loss value threshold, and the image difference degree is less than the preset difference degree Threshold, complete the model training of the mesh removal generator G(x) to obtain the mesh removal model.

Further, the loss value calculation module includes:

Use G(x) to process the image a'obtained by processing the textured image sample a, use F(x) to process the image a" obtained from a', and use F(x) to process the non-textured image sample b to obtain Calculate the first loss value based on the formula (1) based on the image b′ obtained by processing b′ with F(x):

Lg=-(log ₁₀ (Dg(G(a)))-log ₁₀ (Df(F(b)))+Lcyc,

Lcyc=L1Loss(a″,a)×lambda_a+L1Loss(b″,b)×lambda_b+

L1 Loss(a,b')×lambda_c+L1 Loss(b,a')×lambda_d (1)

Among them, Lg is the first loss value, L1 Loss(x, y) represents the Euclidean distance of the two images, and lambda_a, lambda_b, lambda_c, and lambda_d represent preset weights.

Further, the loss value calculation module also includes:

Calculate the second loss value Ldg and the third loss value Ldf based on formula (2) and formula (3):

Ldg=-log ₁₀ (Dg(G(a))-0.5)+log ₁₀ (1.5-Dg(G(a))) (2)

Ldf=-log ₁₀ (Df(F(b))-0.5)+log ₁₀ (1.5-Df(F(b))) (3)

Further, the difference calculation module includes:

The texture image extraction module is used to perform gray value difference calculations on a and b, a and a', and b and b'respectively, and perform a network based on the obtained gray value difference and the preset difference threshold. Texture extraction to obtain the corresponding first, second and third texture images, where a'is the image obtained by processing the texture image sample a using G(x), and b'is the use F(x) The image obtained by processing the non-textured image sample b.

An image difference calculation module for calculating the image distance between the first mesh image and the second mesh image, and the image distance between the first mesh image and the third mesh image , And calculate the difference between the distances of the two images to obtain the image difference degree.

Further, the image difference calculation module includes:

Calculate the image distance L based on formula (4):

Wherein, n is the total number of pixels of the first mesh image or the second mesh image, and x _i and y _i are the _ith of the first mesh image or the second mesh image, respectively The pixel value of each pixel.

For the process of each module in the texture image recognition device provided by the embodiment of the present application to realize their respective functions, please refer to the description of the first embodiment shown in FIG. 1 for details, which will not be repeated here.

It should be understood that the size of the sequence number of each step in the foregoing embodiment does not mean the order of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiment of the present application.

It should also be understood that although the terms “first”, “second”, etc. are used in the text in some embodiments of the present application to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, the first table may be named the second table, and similarly, the second table may be named the first table without departing from the scope of the various described embodiments. The first form and the second form are both forms, but they are not the same form.

Fig. 5 is a schematic diagram of a terminal device provided by an embodiment of the present application. As shown in FIG. 5, the terminal device 5 of this embodiment includes a processor 50 and a memory 51, and the memory 51 stores computer-readable instructions 52 that can run on the processor 50. When the processor 50 executes the computer-readable instructions 52, the steps in the foregoing embodiments of the moire image recognition method are implemented, such as steps 101 to 104 shown in FIG. 1. Alternatively, when the processor 50 executes the computer-readable instructions 52, the functions of the modules/units in the foregoing device embodiments, such as the functions of the modules 41 to 44 shown in FIG. 4, are realized.

The terminal device 5 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server. The terminal device may include, but is not limited to, a processor 50 and a memory 51. Those skilled in the art can understand that FIG. 5 is only an example of the terminal device 5, and does not constitute a limitation on the terminal device 5. It may include more or less components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device may also include an input sending device, a network access device, a bus, and the like.

The so-called processor 50 may be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.

The memory 51 may be an internal storage unit of the terminal device 5, such as a hard disk or a memory of the terminal device 5. The memory 51 may also be an external storage device of the terminal device 5, such as a plug-in hard disk equipped on the terminal device 5, a smart memory card (Smart Media Card, SMC), and a Secure Digital (SD) Card, Flash Card, etc. Further, the memory 51 may also include both an internal storage unit of the terminal device 5 and an external storage device. The memory 51 is used to store the computer-readable instructions and other programs and data required by the terminal device. The memory 51 can also be used to temporarily store data that has been sent or will be sent.

In addition, the functional units in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions, which can be stored in a non-volatile computer. In a readable storage medium, when the computer-readable instructions are executed, they may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still implement the foregoing The technical solutions recorded in the examples are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the application, and should be included in Within the scope of protection of this application.

Claims (20)

1. A method for recognizing a textured image, which is characterized in that it comprises:

   Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the mesh in the image;

   Perform gray value difference calculation on the to-be-processed image and the to-be-processed image without texture, and perform texture reconstruction based on the calculated gray value difference and a preset difference threshold to obtain the corresponding texture Graphics

   Counting the number of pixels contained in the texture pattern, and calculating the ratio of the number of pixels to the total number of pixels in the image to be processed; performing graphic matching on the texture pattern based on a preset texture pattern library;

   If the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.
2. The method for recognizing a reticulated image according to claim 1, wherein the training of the reticulated removal model comprises:

   Obtain multiple pairs of reticulated image samples and non-reticulated image samples, where each pair of image samples includes a reticulated image sample and a non-reticulated image sample, and the reticulated image sample and non-reticulated image sample in each pair of image samples There is only a difference in the texture between the texture image samples;

   Construct a texture removal generator G(x), a texture addition generator F(x), let G(x) get the probability that the image belongs to a non-textured image through the discrimination network Dg(x) as Dg(G(x)) , F(x) obtains the probability that the image belongs to the texture image through the discrimination network Df(x) as Df(F(x));

   Use G(x) and F(x) to process the texture image sample a and the non-texture image sample b respectively to obtain the corresponding processed image, and based on a, b, the processed image, Dg(G( a)) and Df(F(b)) calculate the first loss value corresponding to G(x) and F(x), and calculate Dg(x) and Df based on Dg(G(a) and Df(F(b)) (x) The corresponding second loss value and third loss value;

   Calculating a, b and the degree of image difference between the processed images;

   If the second loss value and/or the third loss value is greater than the corresponding preset loss value threshold value, iteratively update Dg(x) and Df(x); if the first loss value is greater than the corresponding The preset loss value threshold of and/or the image difference degree is substantially equal to the preset difference degree threshold, and G(x) and F(x) are updated iteratively;

   If the first loss value, the second loss value, and the third loss value are all less than the corresponding preset loss value threshold, and the image difference degree is less than the preset difference degree threshold, the texture adjustment is completed The model training of the removal generator G(x) is used to obtain the mesh removal model.
3. 2. The texture image recognition method according to claim 2, wherein said using G(x) and F(x) to process texture image sample a and non-texture image sample b respectively to obtain corresponding processing After image, and calculate the first loss value corresponding to G(x) and F(x) based on a, b, the processed image, Dg(G(a)) and Df(F(b)), including:

   Use G(x) to process the image a'obtained by processing the textured image sample a, use F(x) to process the image a'obtained from a', and use F(x) to process the non-textured image sample b to obtain The image b'of, the image b” obtained by processing b'with F(x), calculate the first loss value based on the following formula:

   Lg=-(log10(Dg(G(a)))-log10(Df(F(b)))+Lcyc,

   Lcyc=L1Loss(a”,a)×lambda_a+L1Loss(b”,b)×lambda_b+

   L1 Loss(a,b')×lambda_c+L1 Loss(b,a')×lambda_d

   Among them, Lg is the first loss value, L1 Loss(x, y) represents the Euclidean distance of the two images, and lambda_a, lambda_b, lambda_c, and lambda_d represent preset weights.
4. The reticulated image recognition method of claim 2, wherein the second loss value corresponding to Dg(x) and Df(x) is calculated based on Dg(G(a) and Df(F(b)) And the third loss value, including:

   Calculate the second loss value Ldg and the third loss value Ldf based on the following formula:

   Ldg=-log10(Dg(G(a))-0.5)+log10(1.5-Dg(G(a)))

   Ldf=-log10(Df(F(b))-0.5)+log10(1.5-Df(F(b))).
5. 8. The method for recognizing a textured image according to claim 3, wherein said calculating a, b and the image difference degree between the processed images comprises:

   Perform gray value difference calculations on a and b, a and a', and b and b'respectively, and perform mesh extraction based on the obtained gray value difference and the preset difference threshold to obtain the corresponding first Reticulated image, second reticulated image and third reticulated image, where a'is the image obtained by processing the reticulated image sample a using G(x), and b'is the non-reticulated image using F(x) Image obtained by processing image sample b;

   Calculate the image distance between the first mesh image and the second mesh image, and the image distance between the first mesh image and the third mesh image, and calculate the two The difference between the image distances is used to obtain the image difference degree.
6. 8. The method for recognizing a mesh image according to claim 5, wherein calculating the image distance between the first mesh image and the second mesh image comprises:

   Calculate the image distance L based on the following formula:

   

   Wherein, n is the total number of pixels of the first mesh image or the second mesh image, and x _i and y _i are the _ith of the first mesh image or the second mesh image, respectively The pixel value of each pixel.
7. A reticulated image recognition device, characterized in that it comprises:

   The de-texturing module is used to input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is based on the reticulated image sample and the non-reticulated image sample in advance. The trained model is used to remove the moire in the image;

   The reticulation reconstruction module is used to calculate the gray value difference of the image to be processed and the non-reticulated image to be processed, and perform reticulation based on the calculated gray value difference and a preset difference threshold Reconstruct to get the corresponding net pattern;

   The feature processing module is used to count the number of pixels contained in the texture pattern, and calculate the ratio of the number of pixels to the total number of pixels in the image to be processed; and compare the texture pattern based on a preset texture pattern library Perform graphic matching;

   The mesh discrimination module is configured to determine that the image to be processed is a mesh image if the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully.
8. 8. The texture image recognition device of claim 7, further comprising:

   The sample acquisition module is used to obtain multiple pairs of textured image samples and non-reticulated image samples, where each pair of image samples includes a textured image sample and a non-reticulated image sample, and each pair of image samples There is only a texture difference between the reticulated image sample and the non-reticulated image sample in;

   The generator building module is used to construct the reticulation removal generator G(x), and the reticulation addition generator F(x). Let G(x) obtain the probability that the image belongs to the non-recreation image through the discrimination network Dg(x) as Dg(G(x)), F(x) through the discriminant network Df(x), the probability that the image belongs to the texture image is Df(F(x));

   Loss value calculation module, used to use G(x) and F(x) to process the texture image sample a and the non-texture image sample b respectively to obtain the corresponding processed image, and based on a, b, and the processing After image, Dg(G(a)) and Df(F(b)) calculate the first loss value corresponding to G(x) and F(x), based on Dg(G(a) and Df(F(b)) Calculate the second loss value and the third loss value corresponding to Dg(x) and Df(x);

   A difference calculation module for calculating the image difference degree between a, b and the processed images;

   An iterative update module, configured to iteratively update Dg(x) and Df(x) if the second loss value and/or the third loss value is substantially equal to the corresponding preset loss value threshold; A loss value is greater than the corresponding preset loss value threshold and/or the image difference degree is greater than the preset difference degree threshold, and G(x) and F(x) are updated iteratively;

   A model output module, configured to: if the first loss value, the second loss value, and the third loss value are all less than a corresponding preset loss value threshold, and the image difference degree is less than the preset difference degree Threshold, complete the model training of the mesh removal generator G(x) to obtain the mesh removal model.
9. 8. The texture image recognition device of claim 8, wherein the loss value calculation module comprises:

   Use G(x) to process the image a'obtained by processing the textured image sample a, use F(x) to process the image a'obtained from a', and use F(x) to process the non-textured image sample b to obtain The image b'of, the image b” obtained by processing b'with F(x), calculate the first loss value based on the following formula:

   Lg=-(log ₁₀ (Dg(G(a)))-log ₁₀ (Df(F(b)))+Lcyc,

   Lcyc=L1Loss(a”,a)×lambda_a+L1Loss(b”,b)×lambda_b+

   L1 Loss(a,b')×lambda_c+L1 Loss(b,a')×lambda_d

   Among them, Lg is the first loss value, L1 Loss(x, y) represents the Euclidean distance of the two images, and lambda_a, lambda_b, lambda_c, and lambda_d represent preset weights.
10. 8. The texture image recognition device of claim 8, wherein the loss value calculation module further comprises:

    Calculate the second loss value Ldg and the third loss value Ldf based on the following formula:

    Ldg=-log ₁₀ (Dg(G(a))-0.5)+log ₁₀ (1.5-Dg(G(a)))

    Ldf=-log ₁₀ (Df(F(b))-0.5)+log ₁₀ (1.5-Df(F(b))).
11. 9. The texture image recognition device of claim 9, wherein the iterative update module comprises:

    The texture image extraction module is used to perform gray value difference calculations on a and b, a and a', and b and b'respectively, and perform a network based on the obtained gray value difference and the preset difference threshold. Texture extraction to obtain the corresponding first, second and third texture images, where a'is the image obtained by processing the texture image sample a using G(x), and b'is the use F(x) the image obtained by processing the non-textured image sample b;

    An image difference calculation module for calculating the image distance between the first mesh image and the second mesh image, and the image distance between the first mesh image and the third mesh image , And calculate the difference between the distances of the two images to obtain the image difference degree.
12. 11. The texture image recognition device of claim 11, wherein:

    Calculate the image distance L based on the following formula:

    

    Wherein, n is the total number of pixels of the first mesh image or the second mesh image, and x _i and y _i are the _ith of the first mesh image or the second mesh image, respectively The pixel value of each pixel.
13. A terminal device, characterized in that, the terminal device includes a memory and a processor, the memory stores computer-readable instructions that can run on the processor, and the processor executes the computer-readable instructions When implementing the following steps:

    Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the mesh in the image;

    Perform gray value difference calculation on the to-be-processed image and the to-be-processed image without texture, and perform texture reconstruction based on the calculated gray value difference and a preset difference threshold to obtain the corresponding texture Graphics

    Counting the number of pixels contained in the texture pattern, and calculating the ratio of the number of pixels to the total number of pixels in the image to be processed; performing graphic matching on the texture pattern based on a preset texture pattern library;

    If the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.
14. The terminal device of claim 13, wherein the training of the mesh removal model includes:

    Obtain multiple pairs of reticulated image samples and non-reticulated image samples, where each pair of image samples includes a reticulated image sample and a non-reticulated image sample, and the reticulated image sample and non-reticulated image sample in each pair of image samples There is only a difference in the texture between the texture image samples;

    Construct a texture removal generator G(x), a texture addition generator F(x), let G(x) get the probability that the image belongs to a non-textured image through the discrimination network Dg(x) as Dg(G(x)) , F(x) obtains the probability that the image belongs to the texture image through the discrimination network Df(x) as Df(F(x));

    Use G(x) and F(x) to process the texture image sample a and the non-texture image sample b respectively to obtain the corresponding processed image, and based on a, b, the processed image, Dg(G( a)) and Df(F(b)) calculate the first loss value corresponding to G(x) and F(x), and calculate Dg(x) and Df based on Dg(G(a) and Df(F(b)) (x) The corresponding second loss value and third loss value;

    Calculating a, b and the degree of image difference between the processed images;

    If the second loss value and/or the third loss value is greater than the corresponding preset loss value threshold value, iteratively update Dg(x) and Df(x); if the first loss value is greater than the corresponding The preset loss value threshold of and/or the image difference degree is substantially equal to the preset difference degree threshold, and G(x) and F(x) are updated iteratively;

    If the first loss value, the second loss value, and the third loss value are all less than the corresponding preset loss value threshold, and the image difference degree is less than the preset difference degree threshold, the texture adjustment is completed The model training of the removal generator G(x) is used to obtain the mesh removal model.
15. The terminal device according to claim 14, characterized in that said using G(x) and F(x) to process the textured image sample a and the non-textured image sample b respectively to obtain the corresponding processed image, And calculate the first loss value corresponding to G(x) and F(x) based on a, b, the processed image, Dg(G(a)) and Df(F(b)), including:

    Use G(x) to process the image a'obtained by processing the textured image sample a, use F(x) to process the image a'obtained from a', and use F(x) to process the non-textured image sample b to obtain The image b'of, the image b” obtained by processing b'with F(x), calculate the first loss value based on the following formula:

    Lg=-(log10(Dg(G(a)))-log10(Df(F(b)))+Lcyc,

    Lcyc=L1Loss(a”,a)×lambda_a+L1Loss(b”,b)×lambda_b+

    L1 Loss(a,b')×lambda_c+L1 Loss(b,a')×lambda_d

    Among them, Lg is the first loss value, L1 Loss(x, y) represents the Euclidean distance of the two images, and lambda_a, lambda_b, lambda_c, and lambda_d represent preset weights.
16. The terminal device of claim 14, wherein the second loss value and the third loss value corresponding to Dg(x) and Df(x) are calculated based on Dg(G(a) and Df(F(b)) Loss value, including:

    Calculate the second loss value Ldg and the third loss value Ldf based on the following formula:

    Ldg=-log10(Dg(G(a))-0.5)+log10(1.5-Dg(G(a)))

    Ldf=-log10(Df(F(b))-0.5)+log10(1.5-Df(F(b))).
17. A computer-readable storage medium that stores computer-readable instructions, wherein the computer-readable instructions are executed by at least one processor to implement the following steps:

    Input the image to be processed into a pre-trained reticulation removal model to obtain a non-reticulated image to be processed. The reticulation removal model is a model obtained by pre-training based on the reticulated image sample and the non-reticulated image sample, and is used for Remove the mesh in the image;

    Perform gray value difference calculation on the to-be-processed image and the to-be-processed image without texture, and perform texture reconstruction based on the calculated gray value difference and a preset difference threshold to obtain the corresponding texture Graphics

    Counting the number of pixels contained in the texture pattern, and calculating the ratio of the number of pixels to the total number of pixels in the image to be processed; performing graphic matching on the texture pattern based on a preset texture pattern library;

    If the number of mesh pixels is greater than a preset number threshold, the ratio value is greater than the preset ratio threshold, and the mesh pattern matches successfully, it is determined that the image to be processed is a mesh image.
18. 17. The computer-readable storage medium of claim 17, wherein the training of the mesh removal model comprises:

    Obtain multiple pairs of reticulated image samples and non-reticulated image samples, where each pair of image samples includes a reticulated image sample and a non-reticulated image sample, and the reticulated image sample and non-reticulated image sample in each pair of image samples There is only a difference in the texture between the texture image samples;

    Construct a texture removal generator G(x), a texture addition generator F(x), let G(x) get the probability that the image belongs to a non-textured image through the discrimination network Dg(x) as Dg(G(x)) , F(x) obtains the probability that the image belongs to the texture image through the discrimination network Df(x) as Df(F(x));

    Use G(x) and F(x) to process the texture image sample a and the non-texture image sample b respectively to obtain the corresponding processed image, and based on a, b, the processed image, Dg(G( a)) and Df(F(b)) calculate the first loss value corresponding to G(x) and F(x), and calculate Dg(x) and Df based on Dg(G(a) and Df(F(b)) (x) The corresponding second loss value and third loss value;

    Calculating a, b and the degree of image difference between the processed images;

    If the second loss value and/or the third loss value is greater than the corresponding preset loss value threshold value, iteratively update Dg(x) and Df(x); if the first loss value is greater than the corresponding The preset loss value threshold of and/or the image difference degree is substantially equal to the preset difference degree threshold, and G(x) and F(x) are updated iteratively;

    If the first loss value, the second loss value, and the third loss value are all less than the corresponding preset loss value threshold, and the image difference degree is less than the preset difference degree threshold, the texture adjustment is completed The model training of the removal generator G(x) is used to obtain the mesh removal model.
19. The computer-readable storage medium according to claim 17, wherein said using G(x) and F(x) to process the textured image sample a and the non-textured image sample b respectively to obtain corresponding processing After image, and calculate the first loss value corresponding to G(x) and F(x) based on a, b, the processed image, Dg(G(a)) and Df(F(b)), including:

    Use G(x) to process the image a'obtained by processing the textured image sample a, use F(x) to process the image a'obtained from a', and use F(x) to process the non-textured image sample b to obtain The image b'of, the image b” obtained by processing b'with F(x), calculate the first loss value based on the following formula:

    Lg=-(log10(Dg(G(a)))-log10(Df(F(b)))+Lcyc,

    Lcyc=L1Loss(a”,a)×lambda_a+L1Loss(b”,b)×lambda_b+

    L1 Loss(a,b')×lambda_c+L1 Loss(b,a')×lambda_d

    Among them, Lg is the first loss value, L1 Loss(x, y) represents the Euclidean distance of the two images, and lambda_a, lambda_b, lambda_c, and lambda_d represent preset weights.
20. The computer-readable storage medium of claim 17, wherein the second loss value corresponding to Dg(x) and Df(x) is calculated based on Dg(G(a) and Df(F(b)) And the third loss value, including:

    Calculate the second loss value Ldg and the third loss value Ldf based on the following formula:

    Ldg=-log10(Dg(G(a))-0.5)+log10(1.5-Dg(G(a)))

    Ldf=-log10(Df(F(b))-0.5)+log10(1.5-Df(F(b))).

Images