WO2020151153A1

WO2020151153A1 - Image processing method and apparatus, and computer device and storage medium

Info

Publication number: WO2020151153A1
Application number: PCT/CN2019/089196
Authority: WO
Inventors: 王义文; 张文龙; 王健宗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-01-24
Filing date: 2019-05-30
Publication date: 2020-07-30
Also published as: CN109886279A; CN109886279B

Abstract

An image processing method and apparatus, and a computer device and a storage medium, by which a pixel value calculation result of a finally obtained feature map having a fixed size is relatively accurate. The method section comprises: obtaining a candidate region feature map; dividing the candidate region feature map into NxM small regions having the same size according to preset candidate region pooling parameters; averagely dividing each small region into P sampling regions according to a preset sampling point number P; determining intersected pixel points intersected with each sampling region in the candidate region feature map (S40); determining a pixel value of a central point position of the sampling region according to the intersected pixel points; correspondingly determining the pixel value of each small region according to the pixel value of the central point position of each sampling region corresponding to the small region (S60); and obtaining the candidate region feature map having a fixed size according to the pixel value of each small region (S70).

Description

Image processing method, device, computer equipment and storage medium

Technical field

This application relates to the field of image processing, and in particular to an image processing method, device, computer equipment and storage medium.

This application is based on the application number 201910067366.3 submitted on January 24, 2019, and the name is "Image processing methods, "Devices, Computer Equipment and Storage Media" are based on the Chinese invention patent application and claim priority.

Background technique

In the field of image processing, it is usually necessary to detect and analyze a certain area, which is called target detection. Among the many frameworks in the field of target detection, such as Fast-RCNN, Faster-RCNN, RFCN, and ROI Pooling's role is to pool the corresponding area into a fixed size in the feature map according to the preselected box Region Proposal, that is, the position coordinates of the candidate area Feature map for subsequent classification operations. However, the position information of Region Proposal is obtained by model regression, and the obtained position information is generally in the form of floating-point numbers, and the feature map of the pooling operation requires a fixed size. Therefore, the process of pooling the corresponding region into a fixed-size feature map involves two rounding and quantization operations: (1) Rounding and quantizing the Region Proposal boundary to integer point coordinate values. (2) The boundary area after rounding and quantization is equally divided into N xN units, and the boundary of each unit is rounded and quantized.

However, after two rounding and quantification, the obtained candidate frame has a certain deviation from the initial regression position. This deviation will affect the accuracy of detection or segmentation, which is called a region misalignment problem (misalignment). In the traditional scheme, the above-mentioned integer quantization operation is cancelled, and the image value on the pixel with the floating point coordinate is obtained by the bilinear interpolation method, so as to transform the entire feature aggregation process into a continuous operation. Yes: First calculate the "red cross point" through the bilinear interpolation algorithm, and calculate the red cross point based on the fixed position beside the red cross point, for example, the lower left and lower right pixels of the pixel where the unified red cross point is located However, the pixel at the fixed position next to the red cross point is not necessarily the pixel close to the pixel point where the red cross point is located, so it may cause a certain error in the calculated result.

Summary of the invention

This application provides an image processing method, device, computer equipment, and storage medium, so that the pixel value calculation result of the fixed-size feature map finally obtained is more accurate.

An image processing method, comprising: obtaining a feature map of a candidate region, the feature map of the candidate region is obtained by mapping the candidate region to a feature map, and the feature map is a feature extraction method of an input image through a convolutional neural network It is obtained that the candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm; dividing the candidate area feature map into NxM small areas of equal size according to a preset candidate area pooling parameter, The N and M are positive integers and greater than or equal to 1, and the preset candidate region pooling parameters include a width parameter and a length parameter for pooling processing; each of the small regions is averaged according to the preset number of sampling points P Divide into P sampling areas, where P is a positive integer and greater than or equal to 2; determine the intersection pixel points that intersect the sampling area in the candidate area feature map; according to the candidate area feature map and the The pixel value at the center point of the sampling area is determined by the intersecting pixel points where the sampling area intersects; according to the pixel value at the center point of each sampling area corresponding to the small area, each of the The pixel value of a small area; obtaining a fixed-size feature map of the candidate area according to the pixel value of each small area.

An image processing device includes: a first acquisition module, configured to acquire a feature map of a candidate region, the feature map of the candidate region is obtained by mapping the candidate region to a feature map, and the feature map is obtained through a convolutional neural network The input image is obtained by feature extraction, and the candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm; the first division module is configured to group the candidate area according to preset candidate area pooling parameters The candidate region feature map acquired by the first acquiring module is divided into NxM small regions of the same size, where N and M are positive integers and greater than or equal to 1, and the preset candidate region pooling parameters include those used for pooling Processed width parameters and length parameters; the second division module is configured to divide the small areas divided by each of the first division modules into P sampling areas on average according to the preset number of sampling points P, where P is positive An integer and greater than or equal to 2; a first determining module, configured to determine the intersecting pixels in the candidate region feature map acquired by the first acquiring module and the sampling region divided by the second dividing module The second determining module is configured to determine the pixel value of the center point of the sampling area according to the intersection pixel point that intersects the sampling area in the feature map of the candidate area determined by the first determining module; Third, a determining module, configured to determine the pixel value of each small area corresponding to the pixel value of the center point of each sampling area corresponding to the small area determined by the second determining module; second An obtaining module is configured to obtain a fixed-size feature map of the candidate area according to the pixel value of each small area determined by the third determining module.

A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor. When the processor executes the computer-readable instructions, the following steps are implemented: obtaining candidates The feature map of the region, the feature map of the candidate region is obtained by mapping the candidate region to the feature map, the feature map is obtained by feature extraction of the input image through a convolutional neural network, and the candidate region is based on a preset The target detection algorithm is obtained by detecting the target area of the input image; according to the preset candidate area pooling parameter, the candidate area feature map is divided into NxM small areas of equal size, where N and M are positive integers and greater than Or equal to 1, the preset candidate area pooling parameters include width parameters and length parameters for pooling processing; each small area is divided into P sampling areas on average according to the preset number of sampling points P, Is a positive integer and greater than or equal to 2; determine the intersection pixel point that intersects the sampling area in the candidate area feature map; determine the intersection pixel point that intersects the sampling area in the candidate area feature map The pixel value at the center point of the sampling area; according to the pixel value at the center point of each sampling area corresponding to the small area, the pixel value of each small area is correspondingly determined; The pixel value of the small area obtains the feature map of the candidate area of a fixed size.

One or more non-volatile readable storage media storing computer readable instructions. When the computer readable instructions are executed by one or more processors, the one or more processors perform the following steps: obtaining A feature map of the candidate area, the feature map of the candidate area is obtained by mapping the candidate area to a feature map, the feature map is obtained by feature extraction of an input image through a convolutional neural network, and the candidate area is based on a prediction Suppose that a target detection algorithm performs target region detection on the input image; according to preset candidate region pooling parameters, the candidate region feature map is divided into NxM small regions of equal size, where N and M are positive integers and Greater than or equal to 1, the preset candidate area pooling parameters include a width parameter and a length parameter for pooling processing; each small area is divided into P sampling areas on average according to the preset number of sampling points P, P is a positive integer and greater than or equal to 2; determine the intersection pixel point that intersects the sampling area in the candidate area feature map; determine the intersection pixel point that intersects the sampling area in the candidate area feature map The pixel value at the center point of the sampling area is determined; according to the pixel value at the center point of each sampling area corresponding to the small area, the pixel value of each small area is correspondingly determined; according to each The pixel value of the small area obtains the candidate area feature map of a fixed size.

The details of one or more embodiments of the present application are presented in the following drawings and descriptions, and other features and advantages of the present application will become apparent from the description, drawings and claims.

Description of the drawings

In order to explain the technical solution of the present application more clearly, the following will briefly introduce the drawings that need to be used in the description of the present application. Obviously, the drawings in the following description are only some embodiments of the present application. Ordinary technicians can also obtain other drawings based on these drawings without creative labor.

Fig. 1 is a schematic diagram of a system frame diagram applied by the image processing method in this application;

2 is a schematic flowchart of an embodiment of the image processing method in this application;

3 is a schematic diagram of the sampling area of the feature map of the candidate area in the present application;

4 is a schematic flowchart of another embodiment of the image processing method in this application;

5 is another schematic diagram of the sampling area of the feature map of the candidate area in the present application;

6 is another schematic diagram of the sampling area of the feature map of the candidate area in the present application;

FIG. 7 is a schematic flowchart of another embodiment of the image processing method in this application;

FIG. 8 is a schematic flowchart of another embodiment of the image processing method in this application;

9 is a schematic flowchart of another embodiment of the image processing method in this application;

10 is another schematic diagram of the sampling area of the feature map of the candidate area in the present application;

FIG. 11 is a schematic flowchart of another embodiment of the image processing method in this application;

FIG. 12 is a schematic structural diagram of an embodiment of an image processing device in this application;

Fig. 13 is a schematic structural diagram of an embodiment of a computer device in this application.

detailed description

The technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The embodiment of the application provides an image processing method, which can be applied to the system framework diagram shown in Fig. 1. The server can execute the processing method provided by this image processing method on the input image to obtain a fixed-size feature map after processing, where: The server can be implemented by an independent server or a server cluster composed of multiple servers. The following describes the embodiments of the present application in detail:

In an embodiment, as shown in FIG. 2, an image processing method is provided, including the following steps:

S10: Obtain a feature map of the candidate region, the feature map of the candidate region is obtained by mapping the candidate region to a feature map, the feature map is obtained by feature extraction of the input image through a convolutional neural network, the candidate region It is obtained by performing target area detection on the input image according to a preset target detection algorithm.

The embodiments of this application are applied to Regions with Convolutional Neural Networks (RCNN), including but not limited to Fast RCNN, Faster RCNN, and Region-based Fully Convolutional Networks (Region-based Fully Convolutional Networks, RFCN). It can be understood that in the above-mentioned region-based convolutional neural network, the input image is processed by the convolutional layer, the pooling layer, the region of interest pooling layer (ROI pooling), and the fully connected layer. However, in the embodiment of the present application, after obtaining the candidate region (Region Proposal) of the input image and the feature maps (ie feature maps) of the input image, the ROI pooling layer performs processing.

Among them, the candidate area is to find out the possible position of the target in the input image in advance, that is, the area of interest, and detect the candidate area of the input image by using feature information such as texture, edge, and color in the input image. The candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm. Specifically, in Fast RCNN, the preset target detection algorithm specifically uses selective search to directly perform a selective search on the input image. Carry out the extraction of candidate regions, and in Faster RCNN, the preset target detection algorithm refers to the feature map of the input image after the feature extraction of the convolutional layer, and then through the regional candidate network (region proposal network, PRN) to perform the feature map The specific process of the extraction of the candidate area will not be described here, but it is understandable that the candidate area corresponding to the input image can be obtained through the processing of the aforementioned preset target detection algorithm.

The feature map of the candidate region refers to the image obtained after mapping the candidate region to the feature map corresponding to the input image. It can be understood that the input image passes through the convolutional layer after inputting the convolutional layer of the region-based convolutional neural network. The feature extraction process can obtain the feature map corresponding to the input object. The specific feature extraction process of the convolutional layer will not be repeated here. In the embodiment of this application, the candidate region feature map corresponding to the input image can be obtained. It should be noted that the candidate region feature map referred to in the embodiment of this application generally refers to each candidate region feature map corresponding to the input image. For ease of description, the following describes the image processing method proposed in the embodiment of the present application with a SAR map of the candidate area.

S20: Divide the candidate region feature map into NxM small regions of equal size according to preset candidate region pooling parameters, where N and M are positive integers and greater than or equal to 1, and the preset candidate region pooling parameters Including width parameters and length parameters.

Among them, the preset candidate region pooling parameters are preset parameters in the ROI pooling layer. The above preset candidate region pooling parameters are intended to pool the candidate region feature maps into parameters of fixed-size feature maps, and the specific address includes the width parameter (pooled-h) and length parameter (pooled-w).

Exemplarily, assuming that the size of the input image is 800*800, the convolutional layer of the region-based convolutional neural network uses the VGG16 network, feat-stride=32 (indicating that the input image is reduced to the original image after the convolutional layer is processed 1/32), that is, the feature map corresponding to the input image obtained after VGG16 network processing is 25*25. Assuming that the input image has a candidate area with a size of 665*665, then the candidate area is mapped to the feature map, and the result is The size of the feature map of the candidate area is: 665/32≈20.78, that is, the feature map of the candidate area is: 20.78*20.78. It should be noted that, in the embodiment of the present application, in the ROI pooling layer processing process, for ease of description, two decimal points are usually reserved for floating-point numbers, and two floating-point numbers after the decimal point are reserved for description when floating-point numbers appear. Assuming that the width parameter and length parameter of the preset candidate region pooling parameters are: pooled-h=7, pooled-w=7, then the candidate region feature map is fixed into a 7*7 feature map after processing, in other words In other words, the 20.78*20.78 candidate area mapped on the feature map is divided into 7*7=49 small areas of equal size, and the size of each small area is 20.78/7≈2.97, which means that each small area Both are 2.97*2.97.

It should be noted that in this embodiment of the application, the candidate region feature map is divided into NxM small regions of equal size according to preset candidate region pooling parameters, and the specific sizes of N and M are configured by actual application requirements. , Related to the preset candidate region pooling parameters, the embodiment of this application does not specifically limit it. For example, the above N and M may also be 8 respectively. At this time, the candidate region feature map is fixed to 8*8 after processing The size of the feature map.

S30: Divide each small area into P sampling areas on average according to the preset number of sampling points P, where P is a positive integer and greater than or equal to 2.

In this embodiment of the application, after the candidate area feature map is divided into NxM small areas of equal size according to the preset candidate area pooling parameters, the following processing is performed for each small area: according to the preset number of sampling points P Each of the small areas is divided into P sampling areas on average, where P is a positive integer and greater than or equal to 2, and the sampling areas have the same shape type as the pixel points of the feature map of the candidate area. Wherein, the preset number of sampling points P is the set number of sampling points used to calculate the pixel value of each small area. Illustratively, the preset number of sampling points may be 4, 8, etc., which is not specifically described in the embodiment of this application. limited. As shown above, exemplarily, after the 20.78*20.78 candidate area obtained by mapping on the feature map is divided into 7*7≈49 small areas of the same size, if the preset number of sampling points P is 4, then 49 Each of the small areas of the same size is divided into 4 sampling areas on average.

S40: Determine an intersection pixel point that intersects the sampling area in the feature map of the candidate area.

After step 30, each small area is divided into P sampling areas on average according to the preset number of sampling points P, and then the intersecting pixel points that intersect the sampling area in the feature map of the candidate area are determined. That is, after each sampling area corresponding to each small area is obtained through the processing of step S30, the intersecting pixel points that respectively intersect each sampling area in the feature map of the candidate area are determined. It should be noted that, specifically, the intersecting pixel points that intersect each sampling area can be determined by the coordinate position of each sampling area and the coordinate position of the pixel point of the candidate area SAR map. Exemplarily, it can be understood that, after step S40, the center point positions of the 4 sampling areas corresponding to each small area and the intersection pixel points that intersect the 4 sampling areas corresponding to the small area can be obtained.

Exemplarily, as shown in Fig. 3: Among them, the area shown in Fig. 3 is only a part of the feature map of the candidate area, including 1-16 pixels, and the area where the thick-lined boxes A, B, C and D are located As the sampling area, sampling areas A, B, C, and D constitute a small area. Taking sampling area A as an example, the position of the center point of sampling area A can be determined by the bilinear difference interpolation algorithm, and the intersecting pixels that intersect with the sampling area A, namely

pixels

1, 2, 5, and 6. . For the sampling area corresponding to each small area, the center point position and the intersection pixel point that intersects the sampling area can be determined. For example, for sampling area B, the target pixel points that intersect with sampling area B are 2, 3, and 6. And 7.

S50: Determine the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area.

After determining the position of the center point of the sampling area, and determining the intersection pixel point that intersects the sampling area in the feature map of the candidate area, according to the intersection pixel point that intersects the sampling area in the feature map of the candidate area Point, the pixel value of the center point of the sampling area is determined. As shown in FIG. 3, taking the sampling area A as an example, the pixel value at the center point of the sampling area A can be determined according to the intersecting

pixels

1, 2, 3, and 4 that intersect the sampling area A. It can be understood that based on the same calculation method, the pixel value of the center point position of each sampling area corresponding to each small area can be obtained respectively.

S60: Correspondingly determine the pixel value of each small area according to the pixel value of the center point position of each sampling area corresponding to each small area.

After step S50, the pixel value of the center point position of the sampling area of each small area in the candidate area can be obtained. In the embodiment of the present application, the pixel value of the center point position of each sampling area corresponding to the small area is obtained. The pixel value corresponds to the pixel value of each small area. As shown in Figure 3, the pixel values of the center point positions of the sampling areas A, B, C, and D corresponding to the small areas can be obtained, respectively, and the correspondingly determined in Figure 3, composed of the sampling areas A, B, C, and D The pixel value of a small area. Based on the same calculation method, the pixel value corresponding to each small area can be obtained, which will not be illustrated here.

S70: Obtain a fixed-size feature map of the candidate area according to the pixel value of each small area.

After the pixel value of each small area is correspondingly determined according to the pixel value of the center point position of each sampling area corresponding to each small area, the processing is obtained according to the pixel value of each small area A feature map of the candidate area with a fixed size afterwards. As shown above, exemplarily, after dividing the 20.78*20.78 candidate area mapped on the feature map into 7*7≈49 small areas of the same size, each small area is 2.97*2.97, after the previous steps S10-S70, get the pixel value corresponding to each small area, get the pixel value of 49 small areas, and output a 7*7 candidate area feature map. Therefore, you can output a fixed size 7*7 candidate area Feature map, the fixed-size candidate region feature map can be used for subsequent classification and regression processing of the region-based convolutional neural network. It can be seen that the embodiment of the present application provides an image processing method, which effectively ensures that the corresponding candidate area is pooled into a fixed-size feature map according to the intersection relationship between the sampling area and the pixel, and because the calculation is performed based on the intersection pixel , The calculation result of the pixel value of the final sampling area is more accurate, so that the finally obtained pixel value calculation result of the fixed-size feature map is also more accurate.

It should be noted that, in combination with the foregoing embodiment, according to the size relationship between the sampling area and the pixel point in the feature map of the candidate area, the embodiment of the present application also provides specific information based on the intersection of the candidate area feature map and the sampling area. The method for determining the pixel value of the center point position of the sampling area by the target intersecting pixel point is described below:

In one embodiment, as shown in FIG. 4, before step 50, that is, according to the target intersecting pixel points that intersect the sampling area in the feature map of the candidate area, the center point position of the sampling area is determined Before the pixel value of, the method further includes the following steps:

S80: Determine whether the size of the sampling area is greater than or equal to the size of the pixel of the feature map of the candidate area.

In step S30, after each small area is divided into P sampling areas on average according to the preset number of sampling points P, it is determined whether the sampling area of each small area is greater than or equal to the size of the pixel point in the feature map of the candidate area It can be understood that since the size of each pixel of the input image is the same, the size of each pixel in the feature map of the candidate area obtained by mapping is also the same, and because each sample of the candidate area obtained by the division The sizes of the regions are the same, so it can be judged whether any sample region is greater than or equal to the size of any pixel in the feature map of the candidate region.

S90: If it is determined that the size of the sampling area is greater than or equal to the size of the pixel point of the candidate area image, generate an auxiliary frame corresponding to the sampling area with the center point position as the center, and the shape of the auxiliary frame is the same as The pixels of the feature map of the candidate area are the same, and the size of the auxiliary frame is smaller than or equal to the size of the pixels of the feature map of the candidate area.

When it is determined that the size of the sampling area is greater than or equal to the pixel size of the candidate area image, then the auxiliary frame corresponding to the sampling area is generated with the center point position as the center, that is, for each small area Each sampling area will generate a corresponding auxiliary frame. The shape of the auxiliary frame is the same as the pixel of the feature map of the candidate area, and the size of the auxiliary frame is smaller than or equal to the size of the pixel of the feature map of the candidate area . It should be noted that the bilinear interpolation method may be used in the embodiment of the present application to determine the center point position of each sampling area, and the process of specifically calculating the center point position of each sampling area will not be repeated here.

Exemplarily, please refer to FIG. 5. As can be seen from FIG. 5, the size of the sampling area is greater than or equal to the size of the pixel of the candidate area image (in FIG. 5, the auxiliary frame and the pixel are the same size as an example). Take the sampling area D as an example. In the embodiment of the present application, the auxiliary frame d corresponding to the sampling area D can be generated at the center point of the sampling area D, as shown by the slash box in FIG. 5, where the sampling area D corresponds to The shape of the auxiliary frame d is the same as the pixels of the feature map of the candidate area, and the size of the auxiliary frame d corresponding to the sampling area D is smaller than or equal to the pixels of the feature map of the candidate area. It should be noted that the sampling area D is only taken as an example for description. For each sampling area of each small area in the candidate area feature map, the corresponding auxiliary frame can be generated through the embodiment of the present application.

In step S50, the determining the pixel value of the center point position of the sampling area according to the intersecting pixel points that intersect the sampling area in the feature map of the candidate area specifically includes the following steps:

S50': Determine the pixel value at the center point of the sampling area according to the intersecting pixels that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area.

After the auxiliary frame of each sampling area in the feature map of the candidate area is generated, the center point position of the sampling area is determined according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area and the corresponding auxiliary frame The pixel value.

For ease of understanding, please refer to Figure 5 again. Taking sampling area D as an example, it can be seen that the intersecting pixels that intersect with sampling area D are

pixels

6, 7, 10, and 11, and the auxiliary frame corresponding to sampling area D is auxiliary frame d, the pixel value at the center point of the sampling area D is determined according to the

pixels

6, 7, 10, 11 and the auxiliary frame d.

It should be noted that the other sampling areas of the feature map of the candidate area are determined based on the pixel value calculation method of the center point position of the sampling area D. For example, please refer to Fig. 6. For the sampling area A, it is visible and sampled. The intersecting pixels in area A are

pixels

1, 2, 5, and 6, and the auxiliary frame corresponding to sampling area A is auxiliary frame a. Then the pixel points 1, 2, 5, 6 and auxiliary frame a are used to determine this The calculation methods of the pixel value at the center point of the sampling area a and the pixel values at the center points of other sampling areas are not described here.

In one embodiment, as shown in FIG. 7, in step S50', that is, according to the intersecting pixel points that intersect the sampling area in the feature map of the candidate area, and the auxiliary frame corresponding to the sampling area, Determining the pixel value of the center point of the sampling area specifically includes the following steps:

S51': Obtain the intersection area of the auxiliary frame corresponding to the sampling area and each pixel point in the intersection pixel point.

S52': Determine the first target pixel value corresponding to the sampling area according to the intersection area of the auxiliary frame corresponding to the sampling area and each pixel point, and the pixel value of each pixel point.

S53': Use the first target pixel value corresponding to the sampling area as the pixel value at the center point of the sampling area.

For step S51`, please continue to refer to FIG. 5, taking the sampling area D as an example, it can be seen that the auxiliary area d corresponding to the sampling area D intersects with the intersecting pixels, that is, the

pixels

6, 7, 10, and 11. In this step , The intersecting areas of the auxiliary frame d and the intersecting pixels ₆ , ₇ , ₁₀ , and 11 can be determined, which are respectively denoted as C _6d , C _7d , C _10d and C _11d in this embodiment.

For steps S52`-S53`, taking the sampling area D as an example, after obtaining C _6d , C _7d , C _10d and C _11d , according to C _6d , C _7d , C _10d and C _11d , and the intersecting

pixels

6, 7, 10, and 11 determine the first target pixel value corresponding to the sampling area D, and use the first target pixel value as the pixel value at the center point of the sampling area D. It should be noted that, for other sampling areas of the feature map of the candidate area, the pixel values of the center point positions corresponding to the sampling areas can be obtained through steps S51`-S53`, and the details are not repeated here.

In an embodiment, as shown in FIG. 8, in step S53', that is, according to the intersection area of the auxiliary frame corresponding to the sampling area and the respective pixel, and the pixel value of the respective pixel, Determining the first target pixel value corresponding to the sampling area specifically includes the following steps:

S531': correspondingly calculate the product between the pixel value of each pixel and the intersection area of each pixel.

S532': Add the product between the pixel value of each pixel and the intersection area of each pixel to obtain a first product sum.

S533': Calculate the sum of the intersection area of each pixel to obtain the sum of the area of the first intersecting pixel.

S534': Calculate the quotient between the first product and the area of the first intersecting pixel to obtain the first target pixel value corresponding to the sampling area.

Here, the sampling area D in the feature map of the candidate area is taken as an example to describe the embodiment of the present application:

For step S531`, after obtaining C _6d, C _7d, C _10d and C _11d, respectively, corresponding to the calculated pixel value of each pixel of C _6d, C _7d, C _10d C _11d and the intersection point of the pixel region corresponding to the sample The product between, that is, the product of C _6d and the pixel value A ₆ of pixel 6 is calculated separately, and it is recorded as: A ₆ C _6d ; the product of C _7d and the pixel value A ₇ of pixel 7 is calculated, and it is recorded as ：A ₇ C _7d ; Calculate the product of C _10d and the pixel value A ₁₀ of the pixel point 10, denoted as: A ₁₀ C _10d ; Calculate the product of C _11d and the pixel value A ₁₁ of the pixel point 11, denoted as: A ₁₁ C _11d .

For step S532`, after calculating A ₆ C _6d , A ₇ C _7d , A ₁₀ C _10d and A ₁₁ C _11d , the above products are added to obtain the first product sum, that is, A ₆ C _6d + A ₇ C _7d +A ₁₀ C _10d +A ₁₁ C _11d .

For step S533', the sum of the intersecting areas of the auxiliary frame d corresponding to the sampling area D and the respective pixels, and the sum of the areas of the first intersecting pixel, that is, C _6d + C _7d + C _10d + C _11d .

For step S234', the quotient U _D between the first product sum corresponding to the sampling area D and the first intersecting pixel area sum is calculated as shown in the following formula:

It should be noted that, in the embodiment of the present application, the sampling area D is taken as an example for description. For other sampling areas in the feature map of the candidate area, the calculation can be performed by referring to the first target pixel value calculation method corresponding to D of the sampling area. , I will not repeat it here. In addition, it should be noted that, according to the size relationship between the sampling area and the size of the pixel points in the feature map of the candidate area, the embodiment of the present application also provides specific intersecting pixels based on the target intersecting the sampling area in the feature map of the candidate area. Point, the method for determining the pixel value at the center point of the sampling area:

As shown in FIG. 9, after step S80, that is, after determining whether the size of the sampling area is greater than or equal to the pixel size of the feature map of the candidate area, the image processing method further includes the following steps:

S100: If it is determined that the size of the sampling area is smaller than the pixel size of the feature map of the candidate area, then according to the pixel value of each intersecting pixel that intersects the sampling area in the feature map of the candidate area, and the The intersection area of the sampling area and each of the intersecting pixel points determines the second target pixel value corresponding to the sampling area.

S110: Use the second target pixel value corresponding to the sampling area as the pixel value at the center point of the sampling area.

As mentioned above, the preset sampling points are preset configurations. The size of the sampling area divided into each small area in the candidate area according to the preset number of sampling points P may be smaller than the size of the pixel points in the feature map of the candidate area, for example When the candidate area is a small target, the feature map of the candidate area corresponding to the input image will be relatively small, and the size of each sample area divided may be smaller than the pixel size of the feature map of the candidate area.

Exemplarily, as shown in FIG. 10: it can be seen that the sampling areas A, B, C, and D divided by the small area are smaller than the size of the pixel, then according to each intersection of the candidate area feature map and the sampling area The pixel value of the pixel point and the intersection area between the sampling area and each of the intersecting pixel points determine the second target pixel value corresponding to the sampling area. Taking sampling area B as an example, the intersecting pixels that intersect with sampling area B are

pixels

5 and 6, respectively. The pixel value of pixel ₅ is A ₅ , the pixel value of pixel ₆ is A ₆ , and sampling area B The intersecting areas with

pixels

5 and 6 are denoted as C _5B and C _{6B respectively} , and then the second target pixel value corresponding to sampling area B is determined according to A ₅ , A ₆ , C _5B and C _6B , and sampling area B corresponds to The second target pixel value of is used as the pixel value of the center point of the sampling area B. For other sampling areas whose size is smaller than the pixel size of the feature map of the candidate area, the pixel values at the positions of the heavy center points of the sampling areas can be determined based on the above-mentioned method, which will not be repeated here.

In one embodiment, as shown in FIG. 11, in step S100, if it is determined that the size of the sampling area is smaller than the pixel size of the feature map of the candidate area, the The pixel value of each intersecting pixel point intersected by the sampling area and the intersecting area of the sampling area and each intersecting pixel point are determined to determine the second target pixel value corresponding to the sampling area, which specifically includes the following steps:

S101: Calculate the product of the pixel value of each intersecting pixel and the intersecting area of each intersecting pixel respectively.

S102: Add the pixel value of each intersecting pixel and the product of the intersecting area of each intersecting pixel to obtain a second sum of products.

S103: Calculate the sum of the intersecting areas of the intersecting pixels to obtain a second sum of the areas of the intersecting pixels.

S104: Calculate the quotient between the second product and the area of the second intersecting pixel to obtain a second target pixel value corresponding to the sampling area.

For steps S101-S104, please continue to refer to Fig. 10, taking the sampling area B in the feature map of the candidate area as an example, after obtaining C _5B and C _6B , the pixel value and the pixel value of each pixel in the intersecting pixel are calculated respectively. The product between C _5B and C _6B , that is, the product of C _5B and the pixel value A ₅ of pixel 5 is calculated respectively, and it is recorded as: A ₅ C _5B ; the pixel value A ₆ of C _6B and pixel 6 is calculated The product of is recorded as: A ₆ C _6B .

Taking the sampling area B in the feature map of the candidate area as an example, the product A ₅ C _5B of C _5B and the pixel value A ₅ of pixel ₅ is obtained, and the pixel value A ₆ of C _6B and pixel 6 is calculated. After the products A ₆ C _6B are multiplied, the products are added to obtain the second product sum, that is, A ₅ C _5B + A ₆ C _6B .

The sum of the intersecting areas of the intersecting pixels corresponding to the sampling area B, that is, the sum of the areas of the first intersecting pixels is: C _5B +C _6B .

The quotient between the second product sum and the area of the second intersecting pixel is calculated by specifically sampling the following formula:

It should be noted that in the embodiment of the present application, the sampling area B is taken as an example for description. For other sampling areas in the feature map of the candidate area, the calculation can be performed by referring to the second target pixel value calculation method corresponding to B of the sampling area. , I will not repeat it 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, and the execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

In one embodiment, an image processing device is provided, and the image processing device corresponds to the image processing method in the above-mentioned embodiment one-to-one. As shown in FIG. 12, the image processing apparatus 10 includes a first acquisition module 101, a first division module 102, a second division module 103, a first determination module 104, a second determination module 105, a third determination module 106, and a second Obtaining module 107. The detailed description of each functional module is as follows:

The first acquisition module is used to acquire a feature map of the candidate area, the feature map of the candidate area is obtained by mapping the candidate area to the feature map, and the feature map is obtained by feature extraction of the input image through a convolutional neural network , The candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm; a first division module is configured to combine the data obtained by the first obtaining module according to a preset candidate area pooling parameter The candidate region feature map is divided into NxM small regions of equal size, where N and M are positive integers and greater than or equal to 1, and the preset candidate region pooling parameters include a width parameter and a length parameter for pooling processing; The second dividing module is configured to divide the small areas divided by each of the first dividing modules into P sampling areas on average according to the preset number of sampling points P, where P is a positive integer and greater than or equal to 2; A determining module, configured to determine the intersection pixel points in the candidate region feature map acquired by the first acquiring module and the sampling area divided by the second dividing module; a second determining module, The pixel value of the center point of the sampling area is determined according to the intersecting pixels in the feature map of the candidate area determined by the first determining module and the sampling area; The pixel value of the center point of each sampling area corresponding to the small area determined by the second determining module determines the pixel value of each small area correspondingly; the second acquiring module is configured to determine the pixel value of each small area according to the The pixel value of each small area determined by the third determining module obtains the candidate area feature map of a fixed size.

In an embodiment, the image processing device further includes a fourth determining module and a generating module;

The fourth determining module is configured to determine before the second determining module determines the pixel value at the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area Whether the size of the sampling area is greater than or equal to the pixel size of the feature map of the candidate area; the generating module is configured to determine that the size of the sampling area is greater than or equal to the candidate area if the fourth determining module determines For the size of the pixel points of the image, the auxiliary frame corresponding to the sampling area is generated with the center point position of the sampling area as the center. The shape of the auxiliary frame is the same as the pixel points of the feature map of the candidate area. The size of the frame is smaller than or equal to the pixel size of the feature map of the candidate area; the second determining module is specifically configured to: according to the intersection pixel point of the feature map of the candidate area that intersects the sampling area, and The auxiliary frame corresponding to the sampling area determines the pixel value of the center point position of the sampling area.

In an embodiment, the second determining module is configured to determine the center of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area The pixel value of the point position includes: the second determining module is configured to: obtain the intersection area of the auxiliary frame corresponding to the sampling area and each pixel point in the intersecting pixel; according to the auxiliary frame corresponding to the sampling area The intersection area with each pixel point and the pixel value of each pixel point determine the first target pixel value corresponding to the sampling area; take the first target pixel value corresponding to the sampling area as the The pixel value of the center point of the sampling area.

In an embodiment, the second determining module is configured to determine the sampling area according to the intersection area of the auxiliary frame corresponding to the sampling area and each pixel point and the pixel value of each pixel point The first target pixel value corresponding to the area includes: the second determining module is specifically configured to: respectively correspondingly calculate the product between the pixel value of each pixel and the intersection area of each pixel; Add the product between the pixel value of the pixel and the intersection area of each pixel to obtain a first product sum; calculate the sum of the intersection area of each pixel to obtain the first intersecting pixel area sum; Calculate the quotient between the first product and the area of the first intersecting pixel to obtain the first target pixel value corresponding to the sampling area.

In an embodiment, the image processing device further includes a fifth determining module: the fifth determining module: after determining whether the size of the sampling area is greater than or equal to the pixel size of the feature map of the candidate area If it is determined that the size of the sampling area is smaller than the size of the pixel of the feature map of the candidate area, then the pixel value of each intersecting pixel that intersects the sampling area in the feature map of the candidate area is determined, and the sampling The intersecting area of the area and each of the intersecting pixel points determines the second target pixel value corresponding to the sampling area; taking the second target pixel value corresponding to the sampling area as the pixel value at the center point of the sampling area .

In an embodiment, the fifth determining module is used for the pixel value of each intersecting pixel that intersects the sampling area in the feature map of the candidate area, and the difference between the sampling area and each intersecting pixel. The intersection area determines the second target pixel value corresponding to the sampling area, including: a fifth determining module is configured to: respectively calculate the pixel value of each target intersection pixel point, and the intersection of each target intersection pixel point The product between the areas; adding the pixel value of each intersecting pixel to the product of the intersecting area of each intersecting pixel to obtain a second product sum; calculating the intersection of each intersecting pixel The sum of the areas is used to obtain the area sum of the second intersecting pixel; the quotient between the second product sum and the area of the second intersecting pixel is calculated to obtain the second target pixel value corresponding to the sampling area.

For the specific definition of the image processing device, please refer to the above definition of the image processing method, which will not be repeated here. Each module in the above-mentioned image processing device may be implemented in whole or in part by software, hardware, and a combination thereof. The foregoing modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the foregoing modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 13. The computer equipment includes a processor, a memory, a network interface and a database connected by a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer equipment is used to store feature maps and so on. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer-readable instructions are executed by the processor to realize an image processing method.

In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, and the processor executes the computer-readable The instruction is implemented as follows: Obtain a feature map of the candidate region, the feature map of the candidate region is obtained by mapping the candidate region to the feature map, and the feature map is obtained by feature extraction of the input image through a convolutional neural network, The candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm; dividing the candidate area feature map into NxM small areas of equal size according to preset candidate area pooling parameters, N and M are positive integers and greater than or equal to 1, the preset candidate area pooling parameters include a width parameter and a length parameter used for pooling processing; each of the small areas is divided equally into P sampling areas, where P is a positive integer and greater than or equal to 2; determine the intersection pixel points that intersect the sampling area in the feature map of the candidate area; according to the feature map of the candidate area and the sampling area Determine the pixel value of the center point of the sampling area for the intersecting pixel points; determine each small area corresponding to the pixel value of the center point of each sampling area corresponding to the small area According to the pixel value of each of the small areas, a fixed-size feature map of the candidate area is obtained.

In one embodiment, one or more non-volatile readable storage media storing computer readable instructions are provided. When the computer readable instructions are executed by one or more processors, the one or more Each processor performs the following steps: Obtain a feature map of the candidate region, the feature map of the candidate region is obtained by mapping the candidate region to a feature map, and the feature map is obtained by feature extraction of the input image through a convolutional neural network The candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm; the candidate area feature map is divided into NxM small areas of equal size according to a preset candidate area pooling parameter, so The N and M are positive integers and greater than or equal to 1, the preset candidate area pooling parameters include a width parameter and a length parameter for pooling processing; each of the small areas is divided equally according to the preset number of sampling points P Are P sampling areas, where P is a positive integer and greater than or equal to 2; determine the intersection pixel points that intersect the sampling area in the candidate area feature map; according to the candidate area feature map and the sampling The pixel value at the center point position of the sampling area is determined for the intersecting pixel points where the area intersects; according to the pixel value at the center point position of each sampling area corresponding to the small area, each small area is correspondingly determined. The pixel value of the area; obtaining a fixed-size feature map of the candidate area according to the pixel value of each small area.

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.

Those skilled in the art can clearly understand that for the convenience and conciseness of description, only the division of the above functional units and modules is used as an example. In actual applications, the above functions can be allocated to different functional units, Module completion means dividing the internal structure of the device into different functional units or modules to complete all or part of the functions described above.

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 they 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

An image processing method, characterized by comprising:

Obtain a feature map of the candidate area, the feature map of the candidate area is obtained by mapping the candidate area to the feature map, the feature map is obtained by feature extraction of the input image through a convolutional neural network, and the candidate area is based on A preset target detection algorithm is obtained by performing target area detection on the input image;

According to preset candidate area pooling parameters, the candidate area feature map is divided into NxM small areas of equal size, where N and M are positive integers and greater than or equal to 1, and the preset candidate area pooling parameters include Width and length parameters for pooling processing;

Divide each small area into P sampling areas on average according to the preset number of sampling points P, where P is a positive integer and greater than or equal to 2;

Determining an intersection pixel point that intersects the sampling area in the candidate area feature map;

Determine the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area;

Correspondingly determine the pixel value of each small area according to the pixel value of the center point position of each sampling area corresponding to the small area;

A fixed-size feature map of the candidate area is obtained according to the pixel value of each small area.
The image processing method according to claim 1, wherein the pixel value of the center point of the sampling area is determined according to the intersection pixel point that intersects the sampling area in the feature map of the candidate area. , The method further includes:

Judging whether the size of the sampling area is greater than or equal to the size of the pixel point of the feature map of the candidate area;

If it is determined that the size of the sampling area is greater than or equal to the size of the pixel point of the candidate area image, the auxiliary frame corresponding to the sampling area is generated with the center point position of the sampling area as the center, and the shape of the auxiliary frame Same as the pixels of the feature map of the candidate area, the size of the auxiliary frame is smaller than or equal to the size of the pixels of the feature map of the candidate area;

The determining the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area includes:

The pixel value of the center point of the sampling area is determined according to the intersecting pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area.
3. The image processing method according to claim 2, wherein the sampling is determined based on the intersection pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area. The pixel value at the center point of the area, including:

Acquiring the intersection area of the auxiliary frame corresponding to the sampling area and each pixel in the intersection pixel;

Determine the first target pixel value corresponding to the sampling area according to the intersection area of the auxiliary frame corresponding to the sampling area and the respective pixel points, and the pixel value of each pixel point;

The first target pixel value corresponding to the sampling area is used as the pixel value at the center point of the sampling area.
5. The image processing method according to claim 3, wherein said determining the said area according to the intersection area of the auxiliary frame corresponding to the sampling area and each pixel and the pixel value of each pixel The first target pixel value corresponding to the sampling area includes:

Correspondingly calculate the product between the pixel value of each pixel and the intersection area of each pixel;

Adding the product between the pixel value of each pixel and the intersection area of each pixel to obtain a first product sum;

Calculating the sum of the intersecting area of each pixel to obtain the sum of the area of the first intersecting pixel;

Calculate the quotient between the first product and the area of the first intersecting pixel to obtain the first target pixel value corresponding to the sampling area.
3. The image processing method according to claim 2, wherein after determining whether the size of the sampling area is greater than or equal to the pixel size of the feature map of the candidate area, the method further comprises:

If it is determined that the size of the sampling area is smaller than the pixel size of the feature map of the candidate area, the pixel value of each intersecting pixel that intersects the sampling area in the feature map of the candidate area and the sampling area Determine the second target pixel value corresponding to the sampling area by the intersection area with each of the intersecting pixel points;

The second target pixel value corresponding to the sampling area is taken as the pixel value at the center point of the sampling area.
The image processing method of claim 5, wherein the pixel value of each intersecting pixel point that intersects the sampling area in the candidate area feature map, and the sampling area and each intersecting pixel value The intersection area of the pixel points to determine the second target pixel value corresponding to the sampling area includes:

Correspondingly calculate the product of the pixel value of each target intersecting pixel and the product of the intersecting area of each target intersecting pixel;

Adding the pixel value of each intersecting pixel to the product of the intersecting area of each intersecting pixel to obtain a second product sum;

Calculating the sum of the intersecting areas of the intersecting pixels to obtain a second sum of the areas of the intersecting pixels;

The quotient between the second product and the area of the second intersecting pixel is calculated to obtain the second target pixel value corresponding to the sampling area.
An image processing device, characterized by comprising:

The first acquisition module is used to acquire a feature map of the candidate area, the feature map of the candidate area is obtained by mapping the candidate area to the feature map, and the feature map is obtained by feature extraction of the input image through a convolutional neural network , The candidate area is obtained by performing target area detection on the input image according to a preset target detection algorithm;

The first division module is configured to divide the candidate region feature map acquired by the first acquisition module into NxM small regions of equal size according to preset candidate region pooling parameters, where N and M are positive integers and greater than Or equal to 1, the preset candidate region pooling parameters include a width parameter and a length parameter used for pooling processing;

The second dividing module is configured to divide the small areas divided by each of the first dividing modules into P sampling areas on average according to the preset number of sampling points P, where P is a positive integer and greater than or equal to 2;

A first determining module, configured to determine intersecting pixels in the candidate area feature map acquired by the first acquiring module and the sampling area divided by the second dividing module;

The second determining module is configured to determine the pixel value of the center point of the sampling area according to the intersection pixel point that intersects the sampling area in the feature map of the candidate area determined by the first determining module;

A third determining module, configured to correspondingly determine the pixel value of each small area according to the pixel value of the center point position of each sampling area corresponding to the small area determined by the second determining module;

The second obtaining module is configured to obtain a fixed-size feature map of the candidate area according to the pixel value of each small area determined by the third determining module.
8. The image processing device of claim 7, wherein the image processing device further comprises a fourth determining module and a generating module;

The fourth determining module is configured to determine before the second determining module determines the pixel value at the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area Whether the size of the sampling area is greater than or equal to the size of the pixel point of the feature map of the candidate area;

The generating module is configured to, if the fourth determining module determines that the size of the sampling area is greater than or equal to the size of the pixel point of the candidate area image, generate the center point of the sampling area as the center. The auxiliary frame corresponding to the sampling area, the shape of the auxiliary frame is the same as the pixels of the feature map of the candidate area, and the size of the auxiliary frame is smaller than or equal to the size of the pixels of the feature map of the candidate area;

The second determining module is specifically configured to: determine the position of the center point of the sampling area according to the intersection pixel point that intersects the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area Pixel values.
The image processing device according to claim 8, wherein the second determining module is used to determine the pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary corresponding to the sampling area. Frame, determining the pixel value of the center point of the sampling area, including:

The second determining module is used for:

Acquiring the intersection area of the auxiliary frame corresponding to the sampling area and each pixel in the intersection pixel;

Determine the first target pixel value corresponding to the sampling area according to the intersection area of the auxiliary frame corresponding to the sampling area and the respective pixel points, and the pixel value of each pixel point;

The first target pixel value corresponding to the sampling area is used as the pixel value at the center point of the sampling area.
9. The image processing device according to claim 9, wherein the second determining module is used for the intersection area between the auxiliary frame corresponding to the sampling area and the respective pixel, and the respective pixel To determine the first target pixel value corresponding to the sampling area, including:

The second determining module is specifically configured to:

Correspondingly calculate the product between the pixel value of each pixel and the intersection area of each pixel;

Adding the product between the pixel value of each pixel and the intersection area of each pixel to obtain a first product sum;

Calculating the sum of the intersecting area of each pixel to obtain the sum of the area of the first intersecting pixel;

Calculate the quotient between the first product and the area of the first intersecting pixel to obtain the first target pixel value corresponding to the sampling area.
A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, wherein the processor executes the computer-readable instructions as follows step:

Obtain a feature map of the candidate area, the feature map of the candidate area is obtained by mapping the candidate area to the feature map, the feature map is obtained by feature extraction of the input image through a convolutional neural network, and the candidate area is based on A preset target detection algorithm is obtained by performing target area detection on the input image;

According to preset candidate area pooling parameters, the candidate area feature map is divided into NxM small areas of equal size, where N and M are positive integers and greater than or equal to 1, and the preset candidate area pooling parameters include Width and length parameters for pooling processing;

Divide each small area into P sampling areas on average according to the preset number of sampling points P, where P is a positive integer and greater than or equal to 2;

Determining an intersection pixel point that intersects the sampling area in the candidate area feature map;

Determine the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area;

Correspondingly determine the pixel value of each small area according to the pixel value of the center point position of each sampling area corresponding to the small area;

A fixed-size feature map of the candidate area is obtained according to the pixel value of each small area.
The computer device according to claim 11, wherein before determining the pixel value of the center point position of the sampling area according to the intersection pixel point that intersects the sampling area in the feature map of the candidate area, When the processor executes the computer-readable instruction, the following steps are also performed:

Judging whether the size of the sampling area is greater than or equal to the size of the pixel point of the feature map of the candidate area;

If it is determined that the size of the sampling area is greater than or equal to the size of the pixel point of the candidate area image, the auxiliary frame corresponding to the sampling area is generated with the center point position of the sampling area as the center, and the shape of the auxiliary frame Same as the pixels of the feature map of the candidate area, the size of the auxiliary frame is smaller than or equal to the size of the pixels of the feature map of the candidate area;

The determining the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area includes:

The pixel value at the center point of the sampling area is determined according to the intersecting pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area.
The computer device according to claim 12, wherein the sampling area is determined based on the intersection pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area The pixel value of the center point position, including:

Acquiring the intersection area of the auxiliary frame corresponding to the sampling area and each pixel in the intersection pixel;

Determine the first target pixel value corresponding to the sampling area according to the intersection area of the auxiliary frame corresponding to the sampling area and the respective pixel points, and the pixel value of each pixel point;

The first target pixel value corresponding to the sampling area is used as the pixel value at the center point of the sampling area.
The computer device according to claim 13, wherein the sampling area is determined according to the intersection area of the auxiliary frame corresponding to the sampling area and each pixel point, and the pixel value of each pixel point. The first target pixel value corresponding to the area includes:

Correspondingly calculate the product between the pixel value of each pixel and the intersection area of each pixel;

Adding the product between the pixel value of each pixel and the intersection area of each pixel to obtain a first product sum;

Calculating the sum of the intersecting area of each pixel to obtain the sum of the area of the first intersecting pixel;

Calculate the quotient between the first product and the area of the first intersecting pixel to obtain the first target pixel value corresponding to the sampling area.
The computer device according to claim 12, wherein after determining whether the size of the sampling area is greater than or equal to the size of the pixel point of the feature map of the candidate area, the method further comprises:

If it is determined that the size of the sampling area is smaller than the pixel size of the feature map of the candidate area, the pixel value of each intersecting pixel that intersects the sampling area in the feature map of the candidate area and the sampling area Determine the second target pixel value corresponding to the sampling area by the intersection area with each of the intersecting pixel points;

The second target pixel value corresponding to the sampling area is taken as the pixel value at the center point of the sampling area.
One or more non-volatile readable storage media storing computer readable instructions, wherein when the computer readable instructions are executed by one or more processors, the one or more processors execute The following steps:

Obtain a feature map of the candidate area, the feature map of the candidate area is obtained by mapping the candidate area to the feature map, the feature map is obtained by feature extraction of the input image through a convolutional neural network, and the candidate area is based on A preset target detection algorithm is obtained by performing target area detection on the input image;

According to preset candidate area pooling parameters, the candidate area feature map is divided into NxM small areas of equal size, where N and M are positive integers and greater than or equal to 1, and the preset candidate area pooling parameters include Width and length parameters for pooling processing;

Divide each small area into P sampling areas on average according to the preset number of sampling points P, where P is a positive integer and greater than or equal to 2;

Determining an intersection pixel point that intersects the sampling area in the candidate area feature map;

Determine the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area;

Correspondingly determine the pixel value of each small area according to the pixel value of the center point position of each sampling area corresponding to the small area;

A fixed-size feature map of the candidate area is obtained according to the pixel value of each small area.
The non-volatile readable storage medium according to claim 16, wherein the center point of the sampling area is determined according to the intersection pixel points that intersect the sampling area in the candidate area feature map Before the pixel value of the position, it also includes:

Judging whether the size of the sampling area is greater than or equal to the size of the pixel point of the feature map of the candidate area;

If it is determined that the size of the sampling area is greater than or equal to the size of the pixel point of the candidate area image, the auxiliary frame corresponding to the sampling area is generated with the center point position of the sampling area as the center, and the shape of the auxiliary frame Same as the pixels of the feature map of the candidate area, the size of the auxiliary frame is smaller than or equal to the size of the pixels of the feature map of the candidate area;

The determining the pixel value of the center point position of the sampling area according to the intersection pixel points that intersect the sampling area in the feature map of the candidate area includes:

The pixel value of the center point of the sampling area is determined according to the intersecting pixel points that intersect the sampling area in the feature map of the candidate area and the auxiliary frame corresponding to the sampling area.
The non-volatile readable storage medium according to claim 17, wherein the intersecting pixel points that intersect the sampling area in the candidate area feature map and the auxiliary frame corresponding to the sampling area , Determining the pixel value of the center point of the sampling area includes:

Acquiring the intersection area of the auxiliary frame corresponding to the sampling area and each pixel in the intersection pixel;

Determine the first target pixel value corresponding to the sampling area according to the intersection area of the auxiliary frame corresponding to the sampling area and the respective pixel points, and the pixel value of each pixel point;

The first target pixel value corresponding to the sampling area is used as the pixel value at the center point of the sampling area.
The non-volatile readable storage medium according to claim 18, wherein the intersecting area of the auxiliary frame corresponding to the sampling area and the respective pixels, and the pixel value of the respective pixels , Determining the first target pixel value corresponding to the sampling area includes:

Correspondingly calculate the product between the pixel value of each pixel and the intersection area of each pixel;

Adding the product between the pixel value of each pixel and the intersection area of each pixel to obtain a first product sum;

Calculating the sum of the intersecting area of each pixel to obtain the sum of the area of the first intersecting pixel;

Calculate the quotient between the first product and the area of the first intersecting pixel to obtain the first target pixel value corresponding to the sampling area.
17. The nonvolatile readable storage medium according to claim 17, wherein after determining whether the size of the sampling area is greater than or equal to the size of the pixel of the feature map of the candidate area, the method further comprises:

If it is determined that the size of the sampling area is smaller than the pixel size of the feature map of the candidate area, the pixel value of each intersecting pixel that intersects the sampling area in the feature map of the candidate area and the sampling area Determine the second target pixel value corresponding to the sampling area by the intersection area with each of the intersecting pixel points;

The second target pixel value corresponding to the sampling area is taken as the pixel value at the center point of the sampling area.