An image edge blur processing system based on deep learning

Abstract

The present invention belongs to the field of image processing technology, and specifically is an image edge blur processing system based on deep learning, comprising: an image input end, an image segmentation module, an image recognition module, a noise reduction processing module, a deep learning subsystem, an image synthesis module and an image output end, wherein the output end of the image input end is connected to the image segmentation module and the image recognition module. The image with blurred edges to be processed is segmented, and the image after the initial segmentation is directly taken out. Based on the image after the initial segmentation, the edge area of ​​the image can be quickly obtained. Since the blurred position is mainly in the edge area of ​​the image, such processing is more targeted. The image after the initial segmentation is further segmented into a plurality of edge patches. By segmenting into a plurality of edge patches, the noise points of one or more edge patches are identified and denoised each time, and there is no need to use multiple images in the same position. Moreover, the processing of the edge patches makes the restored degree of the processed images higher.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an image edge blur processing system based on deep learning. Background Art

[0002] During the imaging process, the edges of the image may be blurred due to focusing, noise, jitter, etc.

[0003] The current processing method is mainly image noise reduction processing to remove image noise. Noise is an important cause of image interference. In practical applications, an image may have various noises, which may be generated during transmission or during quantization. According to the relationship between noise and signal, it can be divided into three forms: (f(x,y) represents the given original image, g(x,y) represents the image signal, and n(x,y) represents the noise.) 1) Additive noise. This type of noise has nothing to do with the input image signal. The noisy image can be expressed as f(x,y)=g(x,y)+n(x,y). Channel noise and the noise generated when the camera of the optical camera tube scans the image belong to this type of noise; 2) Multiplicative noise: This type of noise is related to the image signal. The noisy image can be expressed as f(x,y)=g(x,y)+n(x,y)g(x,y). The noise when the flying spot scanner scans the image, the coherent noise in the TV image, and the granular noise in the film all belong to this type of noise.

[0004] 3) Quantization noise: This type of noise has nothing to do with the input image signal. It is caused by quantization error in the quantization process, which is reflected at the receiving end.

[0005] Currently, the main method for image noise reduction is to directly take multiple images at the same position, merge the images in the blurred noise area and take the average to reduce the noise. In the case of a lack of multiple images at the same position, the current processing method is difficult to deal with the noise in a targeted manner. Summary of the invention

[0006] The purpose of the present invention is to provide an image edge blur processing system based on deep learning to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: an image edge blur processing system based on deep learning, comprising: Image input terminal, image segmentation module, image recognition module, noise reduction processing module, deep learning subsystem, image synthesis module and image output terminal; The output end of the image input end is connected to the image segmentation module and the image recognition module, and the output end of the image recognition module is connected to the image segmentation module; The output end of the image segmentation module is connected to the noise reduction processing module and the deep learning subsystem, and the output end of the deep learning subsystem is connected to the noise reduction processing module; The output end of the noise reduction processing module is connected to the image synthesis module, and the output end of the image synthesis module is connected to the image output end; The specific processing steps of the image edge blur processing system based on deep learning are as follows: S1: Based on the image input end, the image with blurred edges to be processed is input into the image segmentation module, and based on the image recognition module, the image segmentation module is controlled to segment the image, and the edge area of ​​the image is cut out. The edge area of ​​the segmented image is the image after the initial segmentation; S2: further segmenting the image after the initial segmentation in step S1 based on the image segmentation module, so that the image after the initial segmentation is further segmented into multiple edge patches, and the edge patches are output to the noise reduction processing module and the deep learning subsystem in sequence, and the deep learning subsystem further identifies the edge patches and finds out the noise points; S3: The noise reduction processing module eliminates the noise points found, and then assembles the denoised edge patches in order to obtain a processed edge image. The processed edge image is spliced ​​with the internal area of ​​the image and output.

[0008] Preferably, in step S1, the image recognition module is an image recognition module based on deep learning, and the image recognition module cuts the edge of the image at equal distances, and the cut image edge is ≤ 85% of the total area of ​​the image.

[0009] Preferably, the specific method of dividing the image into a plurality of edge patches after the initial segmentation is: Acquire the image after the initial segmentation, pre-layout the segmentation position on the image after the initial segmentation, and determine the appropriate layout; On the image after the initial segmentation, segmentation is performed based on the typesetting form, and the smallest unit in the segmentation process is the pixel point, that is, the segmentation of the pixel points is not destroyed, and small edge patches are obtained.

[0010] Preferably, the edge patch is rectangular, and the size of the edge patch is 20 pixels*20 pixels. When the edge of the edge patch is irregular, the size of the irregular edge patch is not greater than 20 pixels*20 pixels.

[0011] Preferably, the suitable layout is a layout with the least number of irregular edge patches.

[0012] Preferably, the image segmentation module includes a primary image segmentation unit and a secondary image segmentation unit, wherein the primary image segmentation unit segments the image with blurred edges to be processed into a primary segmented image, and the secondary image segmentation unit segments the primary segmented image into small edge pieces.

[0013] Preferably, the image synthesis module includes an image displacement unit, an image stitching unit and an image generation unit, the output end of the image displacement unit is connected to the image stitching unit, and the output end of the image stitching unit is connected to the image generation unit. After determining the segmentation layout method of the image after the initial segmentation, the edge pieces to be cut corresponding to the layout of the image after the initial segmentation are serialized, and they need to correspond one-to-one with each edge piece, and the corresponding positions of the layout and the serial numbers are recorded and saved. The image displacement unit, the image stitching unit and the image generation unit synthesize the denoised edge pieces based on the layout and the serial numbers.

[0014] Preferably, the deep learning subsystem is based on a white coating as a background, and the edge pieces are placed on the white coating background. The deep learning subsystem detects early noise points on the edge pieces on the white coating background.

[0015] Preferably, when the deep learning subsystem identifies a noise point, it obtains the color of the pixels in the area surrounding the noise point, excludes the noise points in the pixels in the surrounding area, averages the color of the pixels in the surrounding area to obtain an average color pixel, and fills the average color pixel in the identified noise point to perform noise reduction.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The image with blurred edges to be processed is segmented, and the image after the initial segmentation is directly taken out. Based on the image after the initial segmentation, the edge area of ​​the image can be quickly obtained. Since the blurred position is mainly in the edge area of ​​the image, this processing is more targeted.

[0017] The image after the initial segmentation is further segmented into multiple edge patches. By segmenting into multiple edge patches, the noise points of one or more edge patches are identified and reduced each time. There is no need to use multiple pictures at the same position, and the processing of the edge patches makes the processed picture have a higher degree of restoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a system logic block diagram of the present invention; Figure 2 It is a system logic block diagram of the image segmentation module of the present invention; Figure 3 It is a system logic block diagram of the image synthesis module of the present invention; Figure 4 The present invention is a flowchart of the process of using the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] In the description of the present invention, it is necessary to understand that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0021] For example, see Figure 1-4 , the present invention provides a technical solution: an image edge blur processing system based on deep learning, comprising: an image input terminal, an image segmentation module, an image recognition module, a noise reduction processing module, a deep learning subsystem, an image synthesis module and an image output terminal; The output end of the image input end is connected to the image segmentation module and the image recognition module, and the output end of the image recognition module is connected to the image segmentation module; the output end of the image segmentation module is connected to the noise reduction processing module and the deep learning subsystem, and the output end of the deep learning subsystem is connected to the noise reduction processing module; the output end of the noise reduction processing module is connected to the image synthesis module, and the output end of the image synthesis module is connected to the image output end; The specific processing steps of the image edge blur processing system based on deep learning are as follows: S1: Based on the image input end, the image with blurred edges to be processed is input into the image segmentation module, and based on the image recognition module, the image segmentation module is controlled to segment the image, and the edge area of ​​the image is cut out. The edge area of ​​the segmented image is the image after the initial segmentation; S2: further segmenting the image after the initial segmentation in step S1 based on the image segmentation module, so that the image after the initial segmentation is further segmented into multiple edge patches, and the edge patches are output to the noise reduction processing module and the deep learning subsystem in sequence, and the deep learning subsystem further identifies the edge patches and finds out the noise points; S3: The noise reduction processing module eliminates the noise points found, and then assembles the denoised edge patches in order to obtain a processed edge image. The processed edge image is spliced ​​with the internal area of ​​the image and output.

[0022] Analysis of the above content: The purpose of the image recognition module is to recognize the image contour and output the initial segmented image that controls the image segmentation module to segment the image within the contour range.

[0023] Image contour extraction is an important technique in image processing, which is mainly used to identify and extract the boundary information of objects in images. Contour extraction can be implemented in many ways, including using functions in the OpenCV library and edge detection algorithms in MATLAB.

[0024] Contour extraction in OpenCV In OpenCV, contour extraction is usually implemented through the cv::findContours() function. This function can find contours in an image and draw these contours through the cv::drawContours() function. The specific steps include reading the grayscale image, performing binarization processing, extracting contours using the findContours() function, and finally drawing the contours.

[0025] Edge Detection Algorithm in MATLAB In MATLAB, edge detection algorithms such as Robert operator, Prewitt operator, Sobel operator, etc. are often used for contour extraction. These algorithms identify edges by detecting points with obvious brightness changes in the image, thereby extracting the contour of the image.

[0026] The deep learning subsystem is used to identify noise points in edge patches, which are caused by uneven grayscale values ​​in edge patches, resulting in spots of varying brightness and darkness appearing on the screen.

[0027] For example 2, please refer to Figure 1-4 The present invention provides a technical solution based on Example 1: in step S1, the image recognition module is an image recognition module based on deep learning, the image recognition module cuts the edge of the image at equal distances, and the cut image edge is ≤ 85% of the total area of ​​the image.

[0028] Analysis of the above content: Since noise mainly appears in the edge area of ​​the image, image segmentation of the edge area of ​​the image is sufficient.

[0029] For example 3, please refer to Figure 1-4 The present invention provides a technical solution based on the first embodiment: the specific method of dividing the image into a plurality of edge patches after the initial segmentation is: The image after the initial segmentation is acquired, the segmentation positions are pre-arranged on the image after the initial segmentation, and a suitable layout is determined, wherein the suitable layout is a layout form with the least number of irregular edge patches; On the image after the initial segmentation, segmentation is performed based on the typesetting form, and the minimum unit in the segmentation process is pixel, that is, the segmentation of pixels is not destroyed, and edge patches are obtained. The edge patches are rectangular, and the size of the edge patches is 20 pixels * 20 pixels. When the edges of the edge patches are irregular, the size of the irregular edge patches is not greater than 20 pixels * 20 pixels.

[0030] The image synthesis module includes an image displacement unit, an image stitching unit and an image generation unit. The output end of the image displacement unit is connected to the image stitching unit, and the output end of the image stitching unit is connected to the image generation unit. After determining the segmentation layout method of the image after the initial segmentation, the edge pieces to be cut corresponding to the layout of the image after the initial segmentation are serialized, and need to correspond to each edge piece one by one, and the corresponding positions of the layout and the serial number are recorded and saved. The image displacement unit, the image stitching unit and the image generation unit synthesize the denoised edge pieces based on the layout and the serial number.

[0031] Analysis of the above content: For example, if an image is rectangular, take the outer edge of the rectangular image to get the image after the initial segmentation. Then, similar to the layout before laying floor tiles, reflect the layout method on the image after the initial segmentation, intelligently select the appropriate layout method, and perform segmentation based on the layout.

[0032] For example, in the layout, each edge piece is numbered and sorted in the order of 1, 2, 3, etc., each need corresponds to the position of each edge piece one by one, and each edge piece is also edited according to the corresponding needs, so that it can be sorted based on this order during subsequent synthesis.

[0033] 6. According to claim 1, a deep learning-based image edge blur processing system is characterized in that: the image segmentation module includes an image primary segmentation unit and an image secondary segmentation unit, the image primary segmentation unit segments the image with blurred edges to be processed into a primary segmented image, and the image secondary segmentation unit segments the primary segmented image into edge patches.

[0034] Example 4, please refer to Figure 1-4Based on the first embodiment, the present invention provides a technical solution: the deep learning subsystem is based on the white coating as the background, and the edge patch is placed on the white coating background. The deep learning subsystem detects early noise points on the edge patch on the white coating background. When the deep learning subsystem identifies a noise point, the color of the pixel points in the surrounding area of ​​the noise point is obtained, and the noise points in the pixel points in the surrounding area are excluded. The color of the pixel points in the surrounding area is averaged to obtain an average color pixel point, and the average color pixel point is filled in the identified noise point to reduce noise.

[0035] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention; therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is limited by the attached claims rather than the above description. Therefore, it is intended to include all changes within the meaning and scope of the equivalent elements of the claims in the present invention, and any figure marks in the claims should not be regarded as limiting the claims involved.

[0036] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image edge blur processing system based on deep learning, characterized in that: include: Image input terminal, image segmentation module, image recognition module, noise reduction processing module, deep learning subsystem, image synthesis module and image output terminal; The output end of the image input end is connected to the image segmentation module and the image recognition module, and the output end of the image recognition module is connected to the image segmentation module; The output end of the image segmentation module is connected to the noise reduction processing module and the deep learning subsystem, and the output end of the deep learning subsystem is connected to the noise reduction processing module; The output end of the noise reduction processing module is connected to the image synthesis module, and the output end of the image synthesis module is connected to the image output end; The specific processing steps of the image edge blur processing system based on deep learning are as follows: S1: Based on the image input end, the image with blurred edges to be processed is input into the image segmentation module, and based on the image recognition module, the image segmentation module is controlled to segment the image, and the edge area of ​​the image is cut out. The edge area of ​​the segmented image is the image after the initial segmentation; S2: further segmenting the image after the initial segmentation in step S1 based on the image segmentation module, so that the image after the initial segmentation is further segmented into multiple edge patches, and the edge patches are output to the noise reduction processing module and the deep learning subsystem in sequence, and the deep learning subsystem further identifies the edge patches and finds out the noise points; S3: The noise reduction processing module eliminates the noise points found, and then assembles the denoised edge patches in order to obtain a processed edge image. The processed edge image is spliced ​​with the internal area of ​​the image and output.

2. The image edge blur processing system based on deep learning according to claim 1, characterized in that: In step S1, the image recognition module is an image recognition module based on deep learning. The image recognition module cuts the edge of the image at equal distances, and the cut image edge is ≤ 85% of the total area of ​​the image.

3. The image edge blur processing system based on deep learning according to claim 1, characterized in that: The specific method of dividing the image into multiple edge patches after the initial segmentation is: Acquire the image after the initial segmentation, pre-layout the segmentation position on the image after the initial segmentation, and determine the appropriate layout; On the image after the initial segmentation, segmentation is performed based on the typesetting form, and the smallest unit in the segmentation process is the pixel, that is, the segmentation of the pixel is not destroyed, and small edge patches are obtained.

4. The image edge blur processing system based on deep learning according to claim 3, characterized in that: The edge patch is rectangular, and the size of the edge patch is 20 pixels*20 pixels. When the edge of the edge patch is irregular, the size of the irregular edge patch is not greater than 20 pixels*20 pixels.

5. The image edge blur processing system based on deep learning according to claim 4, characterized in that: The appropriate layout is the one with the least amount of irregular edge patches.

6. The image edge blur processing system based on deep learning according to claim 1, characterized in that: The image segmentation module includes an image primary segmentation unit and an image secondary segmentation unit. The image primary segmentation unit segments the image with blurred edges to be processed into a primary segmented image, and the image secondary segmentation unit segments the primary segmented image into edge patches.

7. The image edge blur processing system based on deep learning according to claim 3, characterized in that: The image synthesis module includes an image displacement unit, an image stitching unit and an image generation unit. The output end of the image displacement unit is connected to the image stitching unit, and the output end of the image stitching unit is connected to the image generation unit. After determining the segmentation layout method of the image after the initial segmentation, the edge pieces to be cut corresponding to the layout of the image after the initial segmentation are serialized, and need to correspond to each edge piece one by one, and the corresponding positions of the layout and the serial number are recorded and saved. The image displacement unit, the image stitching unit and the image generation unit synthesize the denoised edge pieces based on the layout and the serial number.

8. The image edge blur processing system based on deep learning according to claim 1, characterized in that: The deep learning subsystem is based on a white coating as the background, and the edge pieces are placed on the white coating background. The deep learning subsystem detects early noise points on the edge pieces on the white coating background.

9. The image edge blur processing system based on deep learning according to claim 8, characterized in that: When the deep learning subsystem identifies a noise point, it obtains the color of the pixels in the area surrounding the noise point, excludes the noise points in the pixels in the surrounding area, averages the color of the pixels in the surrounding area to obtain an average color pixel, and fills the average color pixel in the identified noise point to perform noise reduction.

Images