An image processing method based on deep learning

Abstract

The present invention discloses an image processing method based on deep learning, which belongs to the field of image processing technology and comprises the following steps: S1, collecting an image to be processed, and using a point model of pixel points to construct an overall surface model for the image to be processed; S2, constructing a deep learning network, and using the overall surface model of the image to be processed to correct the deep learning network to obtain a standard deep learning network; S3, using the standard deep learning network to enhance the image to be processed. The present invention can greatly improve the image quality, make the image information more prominent, and restore some of its details and features.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and specifically relates to an image processing method based on deep learning. Background Art

[0002] With the popularity of digital cameras and smartphones, image capture has become increasingly convenient. However, in low-light conditions or due to equipment limitations, images captured often have problems with noise and loss of details. Traditional image denoising and enhancement methods, such as filters and histogram equalization, can improve image quality to a certain extent, but they are often not targeted. In recent years, deep learning has shown powerful performance in the field of image processing and provided new solutions for image enhancement. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a deep learning image processing method.

[0004] The technical solution of the present invention is: an image processing method based on deep learning comprises the following steps:

[0005] S1, collecting the image to be processed, and using the point model of the pixel points to construct the overall surface model for the image to be processed;

[0006] S2, constructing a deep learning network, and using the overall surface model of the image to be processed to correct the deep learning network to obtain a standard deep learning network;

[0007] S3. Use a standard deep learning network to enhance the image to be processed.

[0008] Furthermore, S1 includes the following sub-steps:

[0009] S11, collecting an image to be processed, and constructing a point model for each pixel point of the image to be processed;

[0010] S12, constructing a line model for the pixel point according to the point model of the pixel point;

[0011] S13, constructing an overall surface model for the image to be processed according to the line model of each pixel point in the image to be processed.

[0012] The beneficial effect of the above further scheme is: in the present invention, the point, line and surface models are constructed by considering the similarities of the pixels, the pixels in the rows and columns where the pixels are located, and the clustering of the pixels in the entire image in turn, and the processing range of the pixels is gradually expanded, which can ensure that the pixel value of each pixel is fully calculated, so that the constructed overall surface model is more accurate and applicable.

[0013] Furthermore, in S11, the point model x of the i-th pixel iThe expression is:

[0014] Where b i represents the pixel value of the i-th pixel, b i_like represents the pixel value of the pixel closest to the i-th pixel in Euclidean distance, rand(·) represents the generation of a random number between 0 and 1, Indicates rounding up.

[0015] Euclidean distance is used to calculate the straight-line distance between two pixels in RGB color space.

[0016] Furthermore, in S12, the line model y of the i-th pixel i The expression is:

[0017] Where b i represents the pixel value of the i-th pixel, represents the pixel value of the mth pixel in the row where the i-th pixel is located, M represents the number of pixels in the row where the i-th pixel is located except the i-th pixel, represents the pixel value of the nth pixel in the column where the i-th pixel is located, N represents the number of pixels in the column where the i-th pixel is located except the i-th pixel, x i Represents the point model of the i-th pixel.

[0018] Further, S13 includes the following sub-steps:

[0019] S131, clustering the line model values ​​of all pixels to obtain cluster centers;

[0020] S132, calculating the clustering fuzziness between each pixel point and the multidimensional vector corresponding to the cluster center;

[0021] S133. Construct an overall surface model for the image to be processed according to the clustering fuzziness between each pixel point and the multidimensional vector corresponding to the cluster center.

[0022] The beneficial effect of the above further scheme is that: in the present invention, K-Means clustering is performed on all line model values, and the cluster center is a multidimensional numerical vector. The relationship between each pixel point and the multidimensional vector is used, and the overall surface model is constructed in combination with the position of the pixel point to comprehensively characterize the image to be processed.

[0023] Furthermore, in S132, the clustering fuzziness μ between the i-th pixel and the multidimensional vector corresponding to the cluster center is i The calculation formula is:

[0024] Where b irepresents the pixel value of the i-th pixel, d represents the dimension of the multidimensional vector corresponding to the cluster center, represents the norm operation of the vector, and f represents the multidimensional vector corresponding to the cluster center.

[0025] Furthermore, the expression of the overall surface model z of the image to be processed is:

[0026] ; In the formula, μ1 represents the clustering fuzziness between the first pixel and the multidimensional vector corresponding to the cluster center, μ i represents the clustering fuzziness between the i-th pixel and the multidimensional vector corresponding to the cluster center, μ I represents the clustering fuzziness between the multidimensional vector corresponding to the I-th pixel and the cluster center, u1 represents the horizontal coordinate of the first pixel, v1 represents the vertical coordinate of the first pixel, and u i represents the horizontal coordinate of the i-th pixel, v i Indicates the ordinate of the i-th pixel, u I Indicates the horizontal coordinate of the I-th pixel, u I represents the ordinate of the I-th pixel, and I represents the number of pixels in the image to be processed.

[0027] Furthermore, in S2, the deep learning network includes a convolutional layer, an activation layer, a pooling layer, and a fully connected layer connected in sequence.

[0028] The beneficial effect of the above further scheme is: in the present invention, the convolution layer uses multiple convolution kernels to slide on the image to be processed to obtain a feature map; the activation layer uses the activation function to enhance the expression ability of the deep learning network; the pooling layer is used to reduce the dimension of the feature map and improve the generalization ability of the deep learning network; the connection layer uses the activation function to enhance the feature map and complete the image enhancement processing.

[0029] Furthermore, in S2, the fully connected layer of the deep learning network is modified using the overall surface model of the image to be processed to obtain a standard deep learning network;

[0030] The expression of the output L of the fully connected layer of the standard deep learning network is:

[0031] ; Where W represents the weight vector of the fully connected layer of the standard deep learning network, B represents the bias vector of the fully connected layer of the standard deep learning network, represents the norm operation of the vector, Softmax(·) represents the activation function, Pooling represents the output of the pooling layer, and z represents the overall surface model of the image to be processed.

[0032] The beneficial effects of the present invention are: the image processing method constructs a point, line and surface model for the image to be processed captured by the mobile terminal, brings the overall surface model into the deep learning network, improves and modifies the deep learning network, makes the deep learning network more applicable, and completes the image enhancement processing; the present invention can greatly improve the image quality, make the image information more prominent, and restore some of its details and features. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flowchart of the image processing method based on deep learning. DETAILED DESCRIPTION

[0034] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0035] like Figure 1 As shown, the present invention provides an image processing method based on deep learning, comprising the following steps:

[0036] S1, collecting the image to be processed, and using the point model of the pixel points to construct the overall surface model for the image to be processed;

[0037] S2, constructing a deep learning network, and using the overall surface model of the image to be processed to correct the deep learning network to obtain a standard deep learning network;

[0038] S3. Use a standard deep learning network to enhance the image to be processed.

[0039] In this embodiment of the present invention, S1 includes the following sub-steps:

[0040] S11, collecting an image to be processed, and constructing a point model for each pixel point of the image to be processed;

[0041] S12, constructing a line model for the pixel point according to the point model of the pixel point;

[0042] S13, constructing an overall surface model for the image to be processed according to the line model of each pixel point in the image to be processed.

[0043] In the present invention, point, line and surface models are constructed by considering the similarities of pixels, the pixels in the rows and columns where the pixels are located, and the clustering of pixels in the entire image in turn, and the processing range of the pixels is gradually expanded to ensure that the pixel value of each pixel is fully calculated, so that the constructed overall surface model is more accurate and applicable.

[0044] In the embodiment of the present invention, in S11, the point model x of the i-th pixel i The expression is:

[0045] Where bi represents the pixel value of the i-th pixel, b i_like represents the pixel value of the pixel closest to the i-th pixel in Euclidean distance, rand(·) represents the generation of a random number between 0 and 1, Indicates rounding up.

[0046] Euclidean distance is used to calculate the straight-line distance between two pixels in RGB color space.

[0047] In the embodiment of the present invention, in S12, the line model y of the i-th pixel point i The expression is:

[0048] Where b i represents the pixel value of the i-th pixel, represents the pixel value of the mth pixel in the row where the i-th pixel is located, M represents the number of pixels in the row where the i-th pixel is located except the i-th pixel, represents the pixel value of the nth pixel in the column where the i-th pixel is located, N represents the number of pixels in the column where the i-th pixel is located except the i-th pixel, x i Represents the point model of the i-th pixel.

[0049] In this embodiment of the present invention, S13 includes the following sub-steps:

[0050] S131, clustering the line model values ​​of all pixels to obtain cluster centers;

[0051] S132, calculating the clustering fuzziness between each pixel point and the multidimensional vector corresponding to the cluster center;

[0052] S133. Construct an overall surface model for the image to be processed according to the clustering fuzziness between each pixel point and the multidimensional vector corresponding to the cluster center.

[0053] In the present invention, K-Means clustering is performed on all line model values, and the cluster center is a multidimensional numerical vector. The relationship between each pixel point and the multidimensional vector is used, and the overall surface model is constructed in combination with the position of the pixel point to comprehensively characterize the image to be processed.

[0054] In the embodiment of the present invention, in S132, the clustering fuzziness μ between the i-th pixel point and the multidimensional vector corresponding to the cluster center is i The calculation formula is:

[0055] Where b i represents the pixel value of the i-th pixel, d represents the dimension of the multidimensional vector corresponding to the cluster center, represents the norm operation of the vector, and f represents the multidimensional vector corresponding to the cluster center.

[0056] In the embodiment of the present invention, the expression of the overall surface model z of the image to be processed is:

[0057] ; In the formula, μ1 represents the clustering fuzziness between the first pixel and the multidimensional vector corresponding to the cluster center, μ i represents the clustering fuzziness between the i-th pixel and the multidimensional vector corresponding to the cluster center, μ I represents the clustering fuzziness between the multidimensional vector corresponding to the I-th pixel and the cluster center, u1 represents the horizontal coordinate of the first pixel, v1 represents the vertical coordinate of the first pixel, and u i represents the horizontal coordinate of the i-th pixel, v i Indicates the ordinate of the i-th pixel, u I Indicates the horizontal coordinate of the I-th pixel, u I represents the ordinate of the I-th pixel, and I represents the number of pixels in the image to be processed.

[0058] In an embodiment of the present invention, in S2, the deep learning network includes a convolutional layer, an activation layer, a pooling layer and a fully connected layer connected in sequence.

[0059] In the present invention, the convolution layer uses multiple convolution kernels to slide on the image to be processed to obtain a feature map; the activation layer uses the activation function to enhance the expression ability of the deep learning network; the pooling layer is used to reduce the dimension of the feature map and improve the generalization ability of the deep learning network; the connection layer uses the activation function to enhance the feature map and complete the image enhancement processing.

[0060] In the embodiment of the present invention, in S2, the fully connected layer of the deep learning network is corrected using the overall surface model of the image to be processed to obtain a standard deep learning network;

[0061] The expression of the output L of the fully connected layer of the standard deep learning network is:

[0062] ; Where W represents the weight vector of the fully connected layer of the standard deep learning network, B represents the bias vector of the fully connected layer of the standard deep learning network, represents the norm operation of the vector, Softmax(·) represents the activation function, Pooling represents the output of the pooling layer, and z represents the overall surface model of the image to be processed.

[0063] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. An image processing method based on deep learning, characterized in that: The following steps are involved: S1, collecting the image to be processed, and using the point model of the pixel points to construct the overall surface model for the image to be processed; S2, constructing a deep learning network, and using the overall surface model of the image to be processed to correct the deep learning network to obtain a standard deep learning network; S3, using a standard deep learning network to enhance the image to be processed; The S1 comprises the following sub-steps: S11, collecting an image to be processed, and constructing a point model for each pixel point of the image to be processed; S12, constructing a line model for the pixel point according to the point model of the pixel point; S13, constructing an overall surface model for the image to be processed according to the line model of each pixel point in the image to be processed; In S11, the point model x of the i-th pixel i The expression is: Where b i represents the pixel value of the i-th pixel, b i_like represents the pixel value of the pixel closest to the i-th pixel in Euclidean distance, rand(·) represents the generation of a random number between 0 and 1, Indicates rounding up; In S12, the line model y of the i-th pixel point i The expression is: Where b i represents the pixel value of the i-th pixel, represents the pixel value of the mth pixel in the row where the i-th pixel is located, M represents the number of pixels in the row where the i-th pixel is located except the i-th pixel, represents the pixel value of the nth pixel in the column where the i-th pixel is located, N represents the number of pixels in the column where the i-th pixel is located except the i-th pixel, x i Represents the point model of the i-th pixel; The S13 comprises the following sub-steps: S131, clustering the line model values ​​of all pixels to obtain cluster centers; S132, calculating the clustering fuzziness between each pixel point and the multidimensional vector corresponding to the cluster center; S133, constructing an overall surface model for the image to be processed according to the clustering fuzziness between each pixel point and the multidimensional vector corresponding to the cluster center; In S132, the clustering fuzziness μ between the i-th pixel and the multidimensional vector corresponding to the cluster center i The calculation formula is: Where b i represents the pixel value of the i-th pixel, d represents the dimension of the multidimensional vector corresponding to the cluster center, represents the norm operation of the vector, and f represents the multidimensional vector corresponding to the cluster center; The expression of the overall surface model z of the image to be processed is: ; In the formula, μ1 represents the clustering fuzziness between the first pixel and the multidimensional vector corresponding to the cluster center, μ i represents the clustering fuzziness between the i-th pixel and the multidimensional vector corresponding to the cluster center, μ I represents the clustering fuzziness between the multidimensional vector corresponding to the I-th pixel and the cluster center, u1 represents the horizontal coordinate of the first pixel, v1 represents the vertical coordinate of the first pixel, and u i represents the horizontal coordinate of the i-th pixel, v i Indicates the ordinate of the i-th pixel, u I Indicates the horizontal coordinate of the I-th pixel, u I represents the ordinate of the I-th pixel, where I represents the number of pixels in the image to be processed; In S2, the deep learning network includes a convolutional layer, an activation layer, a pooling layer and a fully connected layer connected in sequence; In S2, the fully connected layer of the deep learning network is corrected using the overall surface model of the image to be processed to obtain a standard deep learning network; The expression of the output L of the fully connected layer of the standard deep learning network is: ; Where W represents the weight vector of the fully connected layer of the standard deep learning network, B represents the bias vector of the fully connected layer of the standard deep learning network, represents the norm operation of the vector, Softmax(·) represents the activation function, Pooling represents the output of the pooling layer, and z represents the overall surface model of the image to be processed.

Images