A method for image interpolation based on difference of gaussians

Abstract

The application discloses a kind of based on Gauss difference's image interpolation method, belong to image interpolation technical field.The method first calculates the Gauss difference of original image and obtains the mid-frequency information of image, and it is superimposed to original image, to reduce the edge blur and jaggy phenomenon brought in subsequent step processing process.Second, the corresponding relationship between scaling image coordinates and preprocessed image coordinates is constructed, and according to the corresponding relationship, interpolation calculation is carried out pixel by pixel.Finally, according to the type of original image, the image after interpolation is output.The image interpolation method disclosed by the application has the advantages of low computational complexity, and the Gauss difference technology and the corresponding interpolation method can reduce the edge blur and jaggy phenomenon brought in the interpolation process, suitable for medical imaging, smart city, meteorological remote sensing and other application fields, with high popularization value.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image interpolation, and particularly relates to an image interpolation method based on Gaussian difference. BACKGROUND

[0002] The digital image interpolation technology is mainly for adjusting the resolution of images, and has a wide application in the fields of medical images, smart cities and meteorological remote sensing. In the image processing process, the resolution of the image is adjusted according to the actual application requirements, and the process of adjustment involves image interpolation. Image interpolation is to use the gray values of the known neighboring pixel points to fit the gray value of the unknown pixel point, and the quality of the interpolation method directly affects the subsequent actual engineering application.

[0003] However, in the near neighbor method, the pixel value to be interpolated is set as the pixel value of the nearest point in the original image, and the interpolated image has the sawtooth phenomenon and the blocking effect. The linear interpolation method will produce serious image detail blur and edge sawtooth phenomenon in the image interpolation process. In order to better maintain the edge details and texture information of the image, many machine learning type interpolation methods have appeared in recent years, such as the image interpolation method based on artificial neural network and the method based on variation and regularization. However, there are some uncontrollable factors in this kind of method, and local image distortion and blur problems will occur in the image interpolation process.

[0004] The main problem existing in the current image interpolation method is the loss of detail information in the interpolation process, which is easy to cause edge blur and sawtooth phenomenon. Therefore, how to realize an image interpolation method with clear texture details is particularly important, which has very important research significance and practical value for the fields of medical images, smart cities and meteorological remote sensing. SUMMARY

[0005] In view of the problems existing in the prior art, the application provides an image interpolation method based on Gaussian difference, and the technical scheme adopted by the application is as follows:

[0006] An image interpolation method based on Gaussian difference, comprising the following steps:

[0007] Step S1: Gaussian difference calculation, extracting the medium frequency feature information of the original image F(x, y), generating a medium frequency feature information image H(x, y), and superimposing it with the original image to obtain a preprocessed image P(x, y), so as to reduce the edge blur and sawtooth phenomenon caused in the subsequent interpolation process, wherein the value range of the variable x is x = 1, 2,..., M, the value range of the variable y is y = 1, 2,..., N, and M and N are the row number and column number of the image respectively;

[0008] Step S2: Coordinate mapping, according to the scale coefficient A of image scaling, the corresponding relationship between the coordinates of the scaled image S(x', y') and the coordinates of the preprocessed image P(x, y) is constructed, the reverse coordinate mapping calculation formula is established, so as to determine the pixel points which need to be interpolated after scaling, wherein x' = 1, 2,..., round(M·A), y' = 1, 2,..., round(N·A), "round()" is a rounding function;

[0009] Step S3: Image interpolation, the specific position of the interpolated pixel point in the original image is determined by the reverse coordinate mapping calculation formula in the previous step, and the pixel value of the image S(x', y') is calculated pixel by pixel, if Sx' and Sy' are both integers, that is, the reverse coordinate mapping result coincides with the corresponding pixel point of the preprocessed image P(x, y), then the pixel value of P(x, y) is used, if not, an interpolation algorithm is used to fit the pixel value of the point;

[0010] Step S4: Image output, according to the type of the original image F(x, y), the interpolated image is output, if F(x, y) is a grayscale image, S(x', y') is directly output, if F(x, y) is a color image, the color image is processed in the form of sub-channel, and the sub-channel interpolation results are combined into a color image for output.

[0011] Preferably, the calculation formula of the Gaussian difference in step S1 is:

[0012] Where "*" is the convolution operator, the parameter τ = 10, and σ = 1.5.

[0013] Preferably, the calculation formula of the image P(x, y) in step S1 is:

[0014] P(x, y) = F(x, y) + δ·H(x, y), wherein the parameter δ = 1.25.

[0015] Preferably, the calculation formula of the reverse coordinate mapping in step S2 is:

[0016] Where Sx' and Sy' are the horizontal and vertical coordinates of the reverse coordinate mapping based on the scale coefficient A, x' and y' are the horizontal and vertical coordinates of the scaled image, And are the coordinate mapping adjustment factors of the horizontal and vertical coordinates, respectively.

[0017] Preferably, the image interpolation calculation formula in step S3 is:

[0018] Wherein S is a sub-region with a coordinate point (Sx', Sy') as the center, and the calculation formula of the weight coefficient is

[0019] Compared with the prior art, the present application has the beneficial effect that the principle is simple, the intermediate frequency information is obtained by Gaussian difference calculation on the original image and is superimposed on the original image, so as to reduce the edge blur and jaggy phenomenon caused in the interpolation process. At the same time, the interpolation method in step S3 has low calculation complexity, is suitable for medical imaging, smart city, meteorological remote sensing and other application fields, and has high popularization value. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The embodiment steps of the present application are shown in the figure. DETAILED DESCRIPTION

[0021] In order to facilitate the technical personnel to understand the technical scheme of the present application, the technical scheme of the present application will be further described in combination with the drawings and examples in the specification.

[0022] The present application provides an image interpolation method based on Gaussian difference, and the technical scheme adopted by the present application is as follows:

[0023] In combination with Figure 1 The specific implementation steps of the image interpolation method based on Gaussian difference are as follows:

[0024] Step S1: Gaussian difference calculation, extract the intermediate frequency feature information of the original image F(x, y), generate an intermediate frequency feature information image H(x, y), and superimpose it with the original image to obtain a preprocessed image P(x, y), so as to reduce the edge blur and jaggy phenomenon caused in the subsequent interpolation process, wherein the value range of variable x is x = 1, 2,..., M, the value range of variable y is y = 1, 2,..., N, and M and N are the row number and column number of the image respectively.

[0025] This step is to facilitate the calculation of the weight coefficient in subsequent step S3 and highlight the proportion of the weight of edge details. The calculation formula of Gaussian difference is:

[0026] Wherein "*" is a convolution operator, parameter tau = 10, and sigma = 1.5.

[0027] The calculation formula of the image P(x, y) is: P(x, y) = F(x, y) + delta * H(x, y), wherein parameter delta = 1.25.

[0028] Step S2: Coordinate mapping, according to the scale coefficient A of image scaling, the corresponding relationship between the coordinates of the scaled image S(x', y') and the coordinates of the preprocessed image P(x, y) is constructed, the reverse coordinate mapping calculation formula is established, so as to determine the pixel points which need to be interpolated after scaling, wherein x'=1, 2,..., round(M·A), y'=1, 2,..., round(N·A), and "round()" is a rounding function.

[0029] The scale coefficient A is greater than 1, which means that the original image needs to be enlarged, and 0

[0030] Wherein Sx' and Sy' are the horizontal coordinate and the vertical coordinate of the reverse coordinate mapping based on the scale coefficient A, x' and y' are the horizontal coordinate and the vertical coordinate of the scaled image, and are the coordinate mapping adjustment factors of the horizontal coordinate and the vertical coordinate, respectively.

[0031] Step S3: Image interpolation, the specific position of the pixel point in the original image after interpolation is determined by the reverse coordinate mapping calculation formula in the above step, and the pixel value of the image S(x', y') is calculated pixel by pixel, if Sx' and Sy' are both integers, that is, the reverse coordinate mapping result coincides with the corresponding pixel point of the preprocessed image P(x, y), then the pixel value of P(x, y) is adopted, if not, then an interpolation algorithm is used to fit the pixel value of the point.

[0032] The specific image interpolation calculation formula is:

[0033] Wherein S is a sub-region with (Sx', Sy') as the center, and the calculation formula of the weight coefficient is In the present application, S can be selected as a 3*3 neighborhood or a 6*6 neighborhood, if the real-time requirement of calculation is high, a 3*3 neighborhood can be selected, and the value range of the corresponding variables i, j, k and l is an integer in [-3, 3].

[0034] Step S4: Image output, according to the type of the original image F(x, y), the image after interpolation is output, if F(x, y) is a gray image, then S(x', y') is directly output, if F(x, y) is a color image, then the color image is operated in the above steps S1-S3 in the form of a split channel, and the split channel interpolation results are combined into a color image for output.

[0035] It should be noted that the above embodiments can be freely combined as needed. The above only describes the preferred embodiments of the present application, and it should be pointed out that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. An image interpolation method based on Gaussian difference, characterized in that: The specific steps are as follows: Step S1: Gaussian difference calculation, extract the mid-frequency feature information of the original image F(x,y), generate the mid-frequency feature information image H(x,y), and superimpose it with the original image to obtain the preprocessed image P(x,y), where the value range of variable x is x=1,2,…,M, the value range of variable y is y=1,2,…,N, and M and N are the number of rows and columns of the image, respectively; The formula for calculating the Gaussian difference in step S1 is as follows: Where "*" is the convolution operator, with parameters τ = 10 and σ = 1.5; Step S2: Coordinate mapping. Based on the scaling factor A, construct the correspondence between the coordinates of the scaled image S(x′,y′) and the coordinates of the preprocessed image P(x,y). Establish the inverse coordinate mapping calculation formula to determine the pixels that need to be interpolated after scaling, where x′=1,2,...,round(M·A), y′=1,2,...,round(N·A), and "round()" is the rounding function. The formula for calculating the reverse coordinate mapping in step S2 is as follows: Where Sx′ and Sy′ are the x and y coordinates of the inverse coordinate mapping based on the scale factor A, respectively, and x′ and y′ are the x and y coordinates of the scaled image, respectively. and These are the coordinate mapping adjustment factors for the horizontal and vertical axes, respectively. Step S3: Image interpolation. The specific position of the interpolated pixel in the original image is determined by the inverse coordinate mapping calculation formula in the previous step. The pixel value of image S(x′,y′) is calculated pixel by pixel. If Sx′ and Sy′ are both integers, that is, the inverse coordinate mapping result coincides with the corresponding pixel in the preprocessed image P(x,y), then the pixel value of P(x,y) is used. If they do not coincide, the interpolation algorithm is used to fit the pixel value of the point. The image interpolation calculation formula in step S3 is as follows: Where S is the sub-region centered at coordinate point (Sx′, Sy′), and the formula for calculating the weight coefficient is: S is a 3×3 neighborhood, and the corresponding variables i, j, k and l take values ​​in the range of integers in [-3, 3]. Step S4: Image output. Output the interpolated image according to the type of the original image F(x,y). If F(x,y) is a grayscale image, output S(x′,y′) directly. If F(x,y) is a color image, perform the above steps S1-S3 separately in the form of separate channels, and merge the interpolation results of the separate channels into a single color image for output.

2. The image interpolation method based on Gaussian difference as described in claim 1, characterized in that: The formula for calculating the image P(x,y) in step S1 is: P(x,y)=F(x,y)+δ·H(x,y), where the parameter δ=1.25.

Images