Embedded filtering method based on controllable side window frame

Abstract

The invention discloses an embedded filtering method based on a controllable side window frame, and relates to the field of image processing technology and pattern recognition, and the method comprises the following steps: generating a group of base filters in any direction through the directional derivative of a Gaussian function; a controllable half window and a quarter window are constructed through logical operation, so that a side window set is formed; determining a final output pixel value according to the minimum deviation between the input image and the potential filtering result; and through T times of iteration, generating a filtered image result. The SSW provided by the invention not only overcomes the limitation that the number of sub-windows in the traditional SW is insufficient, but also can more accurately fit the main body structure of the image, so that the perception ability of the filter to detail information is remarkably improved, and the adaptability to pixel distribution is greatly enhanced. According to the method, after the SSW is embedded into a traditional filter, the de-noising and edge-preserving capability of the SSW can be remarkably improved, and the method has wider applicability.

Description

Technical Field

[0001] The present invention relates to the fields of image processing technology and pattern recognition, and more particularly, to an embedded filtering method based on a steerable side window framework. Background Art

[0002] Filtering technology has been widely applied in fields such as defect analysis, target detection, and image segmentation. Currently, researchers have designed a variety of filters, such as Gaussian filters, morphological filters, rolling guidance filters, non-local filters, and joint filters. Although these filters have their own advantages, they all adopt the full window (FW) strategy, so there are certain deficiencies in boundary preservation.

[0003] To solve the above problems, the prior art presents a side window filtering (SW) scheme. This method divides the FW into eight specific sub-windows to measure local similarity and screens potential outputs through a loss function. Different from the traditional FW that places the target pixel at the center, SW places each target pixel at the edge or corner of the window, thus avoiding the boundary crossing of the pixels within the window. Therefore, SW can find the optimal sub-window that contains the target pixel but does not contain the boundary pixel, thereby having stronger edge preservation ability and being able to effectively remove noise. However, due to the usually irregular distribution of image content, the eight fixed windows preset by SW are difficult to adapt to diverse pixel distributions, restricting the detail capture ability of SW. Summary of the Invention

[0004] To solve the above problems, the object of the present invention is to provide an embedded filtering method based on a steerable side window (SSW) framework, aiming to solve the problem of edge blurring caused by the inability of the eight fixed windows in the traditional side window to cover diverse pixel distributions due to the irregular distribution of image content.

[0005] To achieve the above technical object, the present application provides an embedded filtering method based on a steerable side window framework, including the following steps:

[0006] Generating a set of basis filters in arbitrary directions using the directional derivative of the Gaussian function;

[0007] Constructing controllable half-windows and quarter-windows through logical operations to form a set of side windows;

[0008] Determining the final output pixel value according to the minimum deviation between the input image and the potential filtering result;

[0009] Generating the filtered image result after T iterations.

[0010] Preferably, when generating the side window filter, the directional derivative of the Gaussian function is used to control the rotation angle of the base filter.

[0011] Preferably, when generating the set of controllable side windows, a half window is generated through the base filter, and two half windows with different angles are used to synthesize a quarter window. Among them, the side window set composed of the half window and the quarter window adapts to the edge structure of the image to weaken the boundary crossing.

[0012] Preferably, when obtaining the minimum deviation between the input image and the potential filtering result, the minimum deviation is expressed as:

[0013]

[0014] In the formula, e i represents the minimum deviation, x i is the i-th pixel of the input image, represents the pixel set of the n-th side window, represents the adjacent pixels within the n-th side window, represents the kernel weight of the n-th side window, N n is the normalization factor

[0015] Preferably, when determining the final output pixel value, the target pixel is updated according to the side window position corresponding to the minimum deviation.

[0016] Preferably, when performing T iterations, a step-by-step optimization method is adopted for T iterations.

[0017] The present invention provides an embedded filtering system based on a controllable side window framework, including:

[0018] A generation module, configured to generate a set of base filters in any direction by using the directional derivative of the Gaussian function;

[0019] A construction module, configured to construct controllable half windows and quarter windows through logical operations to form a side window set;

[0020] A processing module, configured to determine the final output pixel value according to the minimum deviation between the input image and the potential filtering result;

[0021] An iterative generation module, configured to generate a filtered image result after T iterations.

[0022] Preferably, the generation module is further configured to use the directional derivative of the Gaussian function to control the rotation angle of the base filter.

[0023] Preferably, the construction module is further configured to generate a half window through a base filter, and synthesize a quarter window by using two half windows with different angles, wherein a set of side windows composed of the half window and the quarter window adapts to the edge structure of the image to weaken boundary crossing.

[0024] Preferably, the processing module is further configured to update the target pixel according to the side window position corresponding to the minimum deviation.

[0025] The present invention discloses the following technical effects:

[0026] The present invention uses the derivative of the Gaussian function and logical operations to generate a set of side window sets with controllable properties, which can accurately capture the attribute intensity of target pixels, thereby effectively avoiding the smoothing of boundary pixels. Compared with SW, the present invention is more flexible in sub-window design.

[0027] The present invention can be embedded into various traditional filters, significantly optimizing their local similarity measurement methods and greatly enhancing the edge-preserving ability of images. Compared with SW, the present invention has more advantages in improving the performance of traditional filters. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0029] Figure 1 is the flowchart of the present invention;

[0030] Figure 2 is the rotation schematic diagram of SSW of the present invention;

[0031] Figure 3 is the shape of the box filter under different window strategies of the present invention, where the window radius r = 3. In the figure, (a) is the shape of FW, (b) is the shape of SW, and (c) is the shape of SSW when the interval angle α = 45°;

[0032] Figure 4 is the comparison of the smoothing effects of different window strategies on multiple edge distribution images of the present invention;

[0033] Figure 5 is the comparison of visual edge preservation and noise smoothing in different filter versions of the present invention, where the window radius is r = 3 and the maximum number of iterations is T = 5;

[0034] Figure 6 is the comparison of performance results in different filter versions of the present invention;

[0035] Figure 7 It is a comparison of visual effects on different filter versions of the present invention, where the window radius is r = 5 and the maximum number of iterations is T = 2. Specific implementation manner

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Usually, the components of the embodiments of the present application described and illustrated herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application to be protected, but only represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0037] As Figures 1-7 shown, the present invention provides an embedded filtering method based on a controllable side window framework. By proposing a novel and flexible SSW framework to solve the problem that since the image content often presents an irregular distribution, the eight fixed windows adopted by SW cannot cover the possible pixel distributions. In the framework designed by the present invention, first, a set of basis filters in arbitrary directions is generated by using the directional derivative of the Gaussian function; secondly, a controllable half window and quarter window are constructed through logical operations to form a set of side windows; then, the final output pixel value is determined according to the minimum deviation between the input image and the potential filtering result; finally, after T iterations, the filtered image result is generated, which specifically includes the following processes:

[0038] (1) Input the target pixel from the test image.

[0039] (2) Initialize the parameters, set the window radius, interval angle, and maximum number of iterations.

[0040] (3) Use the directional derivative of the Gaussian function and logical operations to construct a set of controllable side windows composed of half windows and quarter windows.

[0041] (4) Calculate the minimum deviation between the input image and the potential side window filtered pixels to determine the side window position.

[0042] (5) Screen the optimal pixel value according to the minimum deviation.

[0043] (6) Adopt a step-by-step optimization method to iterate steps (4)-(5) and output the filtering result.

[0044] Embodiment: The present invention proposes a novel and flexible SSW framework, which can generate a controllable side window set composed of semi-windows and quarter-windows, effectively improving the perception ability of detailed information and significantly enhancing the adaptability to pixel distribution. Compared with the original SW, SSW makes full use of more sub-window information, not only enhancing the boundary protection ability but also improving the noise suppression effect. In addition, SSW can be flexibly embedded with various filters, not only inheriting the excellent characteristics of SW but also further expanding its flexibility and application scope. The specific process is as follows:

[0045] (1) Input the target pixel x from the test image i .

[0046] (2) Initialize the parameters, set the window radius r, the interval angle α, and the maximum number of iterations T.

[0047] (3) Generate a set of side windows composed of semi-windows with arbitrary angles θ and quarter-windows by using the derivative of the Gaussian function and logical operations and quarter-windows .

[0048] Use the directional derivative of the Gaussian function to flexibly control the rotation angle of the side window, where the Gaussian function can be expressed as:

[0049]

[0050] where σ is the scale factor, x and y are two-dimensional grid coordinates of [-r:r]∈Z, that is, x is a square matrix with 2r + 1 rows, and the values in each row are [-r:r]; y is a square matrix with 2r + 1 columns, and the values in each column are [-r:r].

[0051] Assume G n represents the nth-order derivative of the Gaussian function G with respect to the x direction, (·) θ represents the rotation angle, where θ = 0°:α:(360° - α). That is is obtained by rotating G n by θ angle, then the first-order derivative of the Gaussian function along the x direction can be defined as:

[0052]

[0053] When is rotated by 90 degrees, it can be obtained:

[0054]

[0055] Through and linear combination, a basis filter with an arbitrary angle θ can be constructed, and its expression is:

[0056]

[0057] in, and called The base filter of θ is cos(θ) and sin(θ) are interpolation functions for the base filter. It is not difficult to find that the function Depends on three parameters: rotation angle θ, scale factor σ, and window radius r. At the same time, The generated matrix contains three types of values: zero, negative, and positive samples, where negative and positive samples are symmetrical. Too sparse, set σ = r. Therefore, will be controlled only by the rotation angle θ and the window radius r.

[0058] According to the basic filter The half-window set can be defined as:

[0059]

[0060] Among them, τ represents the minimum value that tends to 0 to the left, that is, τ→0 - According to formula (5), we can get a set of half windows including the center of the circle, such as Figure 2 (a) shown.

[0061] Using two half windows at different angles, a quarter window can be synthesized. The calculation formula is as follows:

[0062]

[0063] Where ∧ represents a logical AND operation. There are only two parameters involved, θ and r. Figure 2 (b) is the rotated quarter window. Figure 2 (c) shows the difference between full windows and controllable side windows (half windows and quarter windows). Figure 2 (d) It can be seen that compared with the full window, the controllable side window can better adapt to the edge structure of the image, thereby weakening the boundary crossing.

[0064] (4) The minimum deviation e between the input image and the potential side window filtered pixels i , determine the optimal side window position, the calculation formula is as follows:

[0065]

[0066] Among them, x i is the i-th pixel of the input image, represents the pixel set of the nth side window, Denote adjacent pixels within the n-th side window, Denote the kernel weights of the n-th side window, and is a set of half windows and quarter windows. N n is the normalization factor, i.e.:

[0067]

[0068] (5) According to e i the corresponding side window position, update the target pixel, and the calculation formula is as follows:

[0069]

[0070] (6) Adopt a step-by-step optimization method. Let x i = ξ i , repeat the iterative steps 4 - 5 for T times to ensure the continuity and smoothness of the filtering process.

[0071] The effects of the present invention are further illustrated by the following experiments:

[0072] Embed the SSW technology of the present invention into a traditional filter and conduct a test run on a workstation configured with an Intel Core(TM) i7 - 6700, 3.4GHz CPU processor and 16GB of memory. Figure 3 Shows the shape differences of different windows under the BoxFilter. From Figure 3 it can be seen that when α = 45°, the SSW can better fit the boundary distribution of the Lena image, so that the SSW can efficiently capture the subtle differences in the image and achieve an efficient smoothing effect.

[0073] To verify the advantages of the SSW technology in edge preservation compared to SW and FW, SSW, SW, and FW are respectively embedded into the box filter for experimental comparison. For ease of description, the full window box filter, side window box filter, and controllable side window box filter are abbreviated as FW - BOX, SW - BOX, and SSW - BOX respectively. Figure 4 Shows the comparison of the smoothing effects of different window strategies on images with various edge distributions. Among them, the second column and the fifth column respectively show the numerical distributions on the median line of different filtering results, and the third column and the sixth column show the corresponding local enlarged details. From Figure 4 it can be seen that the SSW of the present invention can better fit the original data distribution and shows significant advantages in edge preservation.

[0074] To quantify the effectiveness of the SSW filter, it is embedded in different versions of filters. For simplicity of expression, the box filter, median filter, Gaussian filter, bilateral filter, and guided filter are abbreviated as BOX, MED, GAU, BIL, and GUI respectively. In the numerical experiment, there are no additional parameters in BOX and MED except for the parameters r and T. For GAU, its variance is set to 1.5. For BIL, its range similarity and spatial similarity are set to 1.5 and 3 respectively. For GUI, its regularization factor is set to 0.01. It should be noted that when the interval angle α = 90°, SSW is equivalent to SW, and SSW will be adopted subsequently. α＝90° to represent SW. Figure 5 are the joint filtering results of different windows and filtering methods. From Figure 5 it can be seen that when SSW is embedded in different filters, the detail protection ability of the original filter can be improved. And as α decreases, the filtering performance based on SSW will continue to improve. Figure 6 The quantization metrics such as structural similarity (SSIM), feature similarity index measure (FSIM), feature similarity based on color information (FSIMc), and peak signal-to-noise ratio (PSNR) also verify the above view.

[0075] Figure 7 shows the visual comparison results on the Urban100 dataset. Tables 1 and 2 list the SSIM and PSNR metrics of different methods on this dataset respectively.

[0076] Table 1

[0077]

[0078] Table 2

[0079]

[0080] From Figure 7 , Table 1, and Table 2, it can be seen that SSW with controllable angle can significantly improve the performance of traditional filters. It can be seen that compared with the original SW, SSW makes full use of more sub-window information, not only enhancing the boundary protection ability but also improving the noise suppression effect. In addition, SSW can be flexibly embedded in various filters, not only inheriting the excellent characteristics of SW but also further expanding its flexibility and application range.

[0081] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0082] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0083] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. An embedded filtering method based on a controllable side window frame, characterized in that It includes the following steps: Generate a set of basis filters in arbitrary directions by using the directional derivative of the Gaussian function; Construct controllable half-windows and quarter-windows through logical operations to form a set of side windows; Determine the final output pixel value according to the minimum deviation between the input image and the potential filtering result; After T iterations, generate the filtered image result.

2. The embedded filtering method based on a controllable side window framework according to claim 1, wherein: When generating the basis filter, use the directional derivative of the Gaussian function to control the rotation angle of the basis filter.

3. The embedded filtering method based on a controllable side window framework according to claim 2, wherein: When generating the set of controllable side windows, generate a half-window through the basis filter, and synthesize a quarter-window by using two half-windows with different angles. Among them, the set of side windows composed of the half-window and the quarter-window adapts to the edge structure of the image to weaken boundary crossing.

4. The embedded filtering method based on a controllable side window framework according to claim 3, wherein: When obtaining the minimum deviation between the input image and the potential filtering result, the minimum deviation is expressed as: where e i represents the minimum deviation, x i is the i-th pixel of the input image, represents the set of pixels of the n-th side window, represents adjacent pixels within the n-th side window, represents the kernel weight of the n-th side window, N n is the normalization factor 5. The embedded filtering method based on a controllable side window framework according to claim 4, wherein: When determining the final output pixel value, update the target pixel according to the side window position corresponding to the minimum deviation.

6. The embedded filtering method based on a controllable side window framework according to claim 5, wherein: When performing T iterations, adopt a step-by-step optimization method and go through T iterations.

7. An embedded filtering system based on a controllable side window frame, characterized in that It includes: A generation module, which is used to generate a set of basis filters in arbitrary directions by using the directional derivative of the Gaussian function; A construction module, which is used to construct controllable half-windows and quarter-windows through logical operations to form a set of side windows; A processing module, which determines the final output pixel value according to the minimum deviation between the input image and the potential filtering result; An iterative generation module, which is used to generate the filtered image result after T iterations.

8. The embedded filtering system based on a controllable side window framework according to claim 7, wherein: The generation module is further used to control the rotation angle of the basis filter by using the directional derivative of the Gaussian function.

9. The embedded filtering system based on a controllable side window framework according to claim 8, wherein: The construction module is further used to generate a half-window through the basis filter, and synthesize a quarter-window by using two half-windows with different angles. Among them, the set of side windows composed of the half-window and the quarter-window adapts to the edge structure of the image to weaken boundary crossing.

10. The embedded filtering system based on a controllable side window framework according to claim 9, wherein: The processing module is further used to update the target pixel according to the side window position corresponding to the minimum deviation.

Images