CN106920222A - A kind of image smoothing method and device - Google Patents

Abstract

A kind of image smoothing method is the embodiment of the invention provides, original image is smoothed first with the bilateral filtering in local smoothing method method and transform domain filtering, obtain navigational figure；Then least square model is constructed to original image, navigational figure and default smoothed image using least square method, adds the constraint function to default smoothed image to control the degree of rarefication of smoothed image, obtain smoothed energy object function；Finally using half secondary split method and alternately, fixed variable method solves the function, so as to obtain original image by the smoothed image after smoothing processing.Consider global characteristics and local feature, enhance the protection of constituent in original image, while having recovered the details of some high-contrasts, the image smoothing effect for obtaining is conducive to improving the accuracy rate of image recognition.Additionally, the embodiment of the present invention is additionally provided realizes device accordingly, further such that methods described has more practicality, described device has corresponding advantage.

Description

Image smoothing method and device

technical field

The embodiments of the present invention relate to the technical field of image processing, and in particular, to an image smoothing method and device.

Background technique

With the rapid development of computer image processing technology, image smoothing technology has also developed rapidly in order to meet the high requirements of the current image processing technology field. Image smoothing technology refers to the image processing technology used to highlight the wide area, low-frequency components, and main parts of the image or suppress image noise and interference high-frequency components, make the image brightness gradually change, reduce the abrupt gradient, and improve image quality. It is widely used In image segmentation, denoising, detail enhancement, object classification, edge extraction and other fields.

In the process of image acquisition and transmission, it is unavoidable to be disturbed by noise and unimportant details (especially high contrast), which makes the target to be recognized appear inconspicuous, which brings difficulties to the recognition. Using image smoothing technology to process images can avoid interference to a certain extent, thereby improving the success rate of image recognition.

In the prior art, image smoothing methods generally include local smoothing methods and global smoothing methods. Local smoothing methods, such as Gaussian filtering, bilateral filtering, median filtering, transform domain filtering, etc., refer to processing local regions or patches of images. The local smoothing method only takes into account the local area characteristics of the image, and has a better smoothing effect on the local area of the image, but it is easy to cause the problem of structural blur. Global smoothing methods, such as total variational smoothing algorithm, weighted least mean square smoothing algorithm, L ₀ gradient minimization smoothing algorithm, etc., process all areas of the entire image at the same time. The structure of the global smoothing method keeps the optimization framework of the constraint items more flexible, and it is better than the local smoothing method for the global features of the image, especially for the unimportant details such as the background part of the image, but the global smoothing algorithm often affects the local high contrast noise. Smoothing is less effective.

In summary, local smoothing methods and global smoothing methods have their own advantages and disadvantages. When smoothing images, how to comprehensively apply global features and local features, avoid the disadvantages of local smoothing methods and global smoothing methods, and improve image smoothing algorithms. Robustness and obtaining a good image smoothing effect are urgent problems to be solved by those skilled in the art.

Contents of the invention

The purpose of the embodiments of the present invention is to provide an image smoothing method and device, so as to improve the robustness of an image smoothing algorithm and obtain a good image smoothing effect.

In order to solve the above technical problems, embodiments of the present invention provide the following technical solutions:

An embodiment of the present invention provides an image smoothing method, including:

Perform preset times of bilateral filtering and transform domain filtering on the original image to obtain a guide image;

Constructing a least squares model for the original image, the guide image and the preset smoothed image according to the least squares method;

defining the pixel intensity and gradient of the preset smooth image by norm, so as to obtain the constraint function of the preset smooth image;

obtaining a smooth energy objective function according to the least squares model and the constraint function;

The smooth energy objective function is solved by semi-quadratic splitting method and alternating fixed variable method to obtain a smooth image.

Optionally, the least squares model constructed for the original image, the guide image and the preset smooth image according to the least squares method is:

According to the least squares method and the square of 2-norm, the least squares model is constructed for the original image, the guide image and the preset smooth image as:

In the formula, S is the preset smooth image, I is the original image, G is the guide image, and α is the detail restoration factor.

Optionally, the definition of pixel intensity and gradient of the preset smoothing image using a norm is:

The 0-norm is used to define the pixel intensity and gradient of the preset smoothing image.

Optionally, the smooth energy objective function obtained according to the least squares model and the constraint function is:

According to the least squares model and the constraint function, the smooth energy objective function is:

In the formula, E(S) is the smoothing energy objective function, S is the preset smoothing image, I is the original image, G is the guiding image, α is the detail restoration factor, and λ is the smoothing factor , ||▽S|| ₀ is the constraint function.

Optionally, the least squares model constructed for the original image, the guide image and the preset smooth image according to the least squares method is:

A least squares model is constructed on the median filtered original image, the guide image and the preset smoothed image according to the least squares method.

Optionally, performing bilateral filtering and transform domain filtering with a preset number of cycles on the original image to obtain a guide image includes:

S1: Obtain the original image, constant value image, spatial weight value and range weight value;

S2: Using the constant value image as an initial guiding function, perform transform domain filtering on the original image and the initial guiding function according to a transform domain filtering method to obtain a new guiding function;

S3: Perform bilateral filtering on the original image and the new guiding function according to the bilateral filtering method to obtain a primary guiding function;

S4: cyclically execute S2 and S3 for the preset number of times to obtain the guide image.

Optionally, the preset number of times is 3 times.

Optionally, the space weight value and range weight value are:

The spatial weight value is 3;

The range weight value is 0.01.

Optionally, solving the smooth energy objective function by using the semi-quadratic splitting method and the alternating fixed variable method to obtain a smooth image includes:

introducing auxiliary variables to replace the constraint function term in the smooth energy objective function;

Using the semi-quadratic splitting method to minimize the replaced smooth energy objective function, adding an error penalty term to obtain a smooth minimization model;

Solving the smoothing minimization model according to the alternating fixed variable method to obtain the smoothed image.

Another aspect of the embodiment of the present invention provides an image smoothing device, including:

A filtering and smoothing module, for performing a preset number of bilateral filtering and transform domain filtering on the original image to obtain a guide image;

Establishing a model module for constructing a least squares model for the original image, the guide image and the preset smooth image according to the least squares method;

Obtaining a smooth image module, used to define the pixel intensity and gradient of the preset smooth image by norm, so as to obtain the constraint function of the preset smooth image; obtain the smooth energy according to the least squares model and the constraint function Objective function: using the semi-quadratic splitting method and the alternating fixed variable method to solve the smooth energy objective function to obtain a smooth image.

The embodiment of the present invention provides an image smoothing method. First, the original image is smoothed by bilateral filtering and transform domain filtering in the local smoothing method to obtain a guide image; then the original image, the guide image and the preset The least squares model is constructed from the smooth image, and the constraint function on the preset smooth image is added to control the sparsity of the smooth image to obtain the smooth energy objective function; finally, the function is solved by using the semi-quadratic splitting method and the alternating fixed variable method to obtain the original A smoothed image after the image has been smoothed.

The technical solution provided by this application comprehensively considers the global features and local features, first performs local smoothing processing on the original image, and then performs global smoothing processing, avoids the disadvantages of the local smoothing method and the global smoothing method, and effectively utilizes the advantages of both . By controlling the difference between the smoothed image and the original image and controlling the difference between the smoothed image and the guided image, the protection of the structural components in the original image is enhanced, the structure of the image is preserved, and some high-contrast details are restored while removing detailed texture features. , to obtain a good image smoothing effect; in addition, effective noise filtering is carried out on the image, which strengthens the intensity of boundary pixels, which is beneficial to the extraction of image contours, and thus helps to improve the accuracy and efficiency of image recognition.

In addition, the embodiment of the present invention also provides a corresponding implementation device for the image smoothing method, which further makes the method more practical, and the device has corresponding advantages.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only For some embodiments of the present invention, those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1-1 is the original image of an exemplary example provided by the embodiment of the present invention;

Fig. 1-2 is the smoothed image of the original image in Fig. 1-1 provided by the embodiment of the present invention;

FIG. 2 is a schematic flowchart of an image smoothing method provided by an embodiment of the present invention;

Fig. 3 is the original image of another exemplary example provided by the embodiment of the present invention;

Fig. 4 is the image obtained by smoothing the original image in Fig. 3 provided by an embodiment of the present invention by inputting different spatial weight values;

FIG. 5 is an image obtained by smoothing the original image in FIG. 3 provided by an embodiment of the present invention by inputting different range weight values;

Fig. 6 is an image obtained by smoothing the original image in Fig. 3 provided by an embodiment of the present invention by inputting different detail restoration factors;

FIG. 7 is an image obtained by smoothing the original image in FIG. 3 provided by an embodiment of the present invention by inputting different smoothing factors;

FIG. 8 is a structural diagram of an implementation manner of an image smoothing device provided by an embodiment of the present invention;

FIG. 9 is an original image of another exemplary example provided by an embodiment of the present invention;

FIG. 10 is an image obtained by smoothing the original image in FIG. 9 provided by an embodiment of the present invention through RGF and BLF algorithms;

Fig. 11 is the image obtained by smoothing the original image in Fig. 9 provided by the embodiment of the present invention through the RTV algorithm;

Fig. 12 is the image obtained by smoothing the original image in Fig. 9 provided by the embodiment of the present invention through the NLGRTV algorithm;

Fig. 13 is the image obtained by smoothing the original image in Fig. 9 provided by the embodiment of the present invention through the SSPTF algorithm;

Fig. 14 is the image obtained by smoothing the original image in Fig. 9 provided by the embodiment of the present invention through the algorithm provided by the present application;

Fig. 15 is an original image of another exemplary example provided by the embodiment of the present invention;

FIG. 16 is an image obtained by denoising the original image in FIG. 15 provided by an embodiment of the present invention through RGF and BLF algorithms;

Fig. 17 is an image obtained by denoising the original image in Fig. 15 provided by the embodiment of the present invention through the algorithm provided by the present application;

Fig. 18 is an original image of another exemplary example provided by the embodiment of the present invention;

FIG. 19 is an image obtained by performing image enhancement processing on the original image in FIG. 19 provided by an embodiment of the present invention through the algorithm provided in this application.

detailed description

In order to enable those skilled in the art to better understand the solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. Apparently, the described embodiments are only some of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The terms "first", "second", "third" and "fourth" in the specification and claims of this application and the above drawings are used to distinguish different objects, rather than to describe a specific order . Furthermore, the terms "comprising" and "having", and any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device comprising a series of steps or units is not limited to the listed steps or units, but may include unlisted steps or units.

The inventors of the present application have found through research that in the prior art, only local features or global features are often considered, resulting in either a blurred structure of the smoothed image, or poorly processed details and large local noise. In short, the image The smoothing effect is difficult to meet the requirements of the technical field of image processing. In view of this, this application first performs local smoothing processing on the original image and then global smoothing processing by comprehensively considering global features and local features, avoiding the disadvantages of local smoothing methods and global smoothing methods, and effectively utilizing the advantages of both. A good image smoothing effect is obtained.

Based on the above-mentioned technical solution of the embodiment of the present invention, some possible application scenarios related to the technical solution of the embodiment of the present invention will be described below with reference to Fig. 1 and Fig. 2. Fig. 1-1 is the original image provided by the embodiment of the present invention. Figures 1-2 are images processed by the method provided in this application.

First use the bilateral filtering and transform domain filtering in the local smoothing method to smooth the original image (Figure 1-1) to obtain the guide image; then use the least squares method to construct the least squares of the original image, the guide image and the preset smoothing image model, adding a constraint function to the preset smooth image to control the sparsity of the smooth image, and obtain the smooth energy objective function; finally, use the semi-quadratic splitting method and the alternating fixed variable method to solve the function, so as to obtain the original image after smoothing Smooth the image (Figure 1-2). Comparing Figure 1-1 and Figure 1-2, it can be seen that the technical solution provided by this application not only preserves the structure of the image after smoothing the graphics, but also restores some high-contrast details while removing detailed texture features. A nice smoothing effect is obtained.

It should be noted that the above application scenarios are only shown for the convenience of understanding the ideas and principles of the present application, and the implementation manners of the present application are not limited in this regard. On the contrary, the embodiments of the present application can be applied to any applicable scene.

After introducing the technical solutions of the embodiments of the present invention, various non-limiting implementations of the present application will be described in detail below.

First, referring to FIG. 2, FIG. 2 is a schematic flowchart of an image smoothing method provided by an embodiment of the present invention. The embodiment of the present invention may include the following:

S201: Circularly perform bilateral filtering and transform domain filtering for a preset number of times on the original image to obtain a guide image.

Both bilateral filtering and transform domain filtering are local smoothing methods.

Bilateral filtering is a nonlinear filtering method, which is a compromise processing combining the spatial proximity of the image and the similarity of pixel values, while considering the spatial information and gray similarity to achieve the purpose of edge preservation and denoising. It is simple, non-iterative, and local. The advantage of bilateral filtering is that it can be used for edge preservation. Generally, Wiener filtering or Gaussian filtering used in the past for noise reduction will blur the edges more obviously, and the protection effect on high-frequency details is not obvious. As the name implies, the bilateral filter has a Gaussian variance more than the Gaussian filter. It is a Gaussian filter function based on the spatial distribution, so near the edge, the pixels farther away will not affect the pixel value on the edge too much, so that it is guaranteed Saving of pixel values near edges. However, due to the preservation of too much high-frequency information, the bilateral filter cannot cleanly filter out the high-frequency noise in the color image, and can only filter the low-frequency information better.

After many experiments, it is found that when only bilateral filtering is used, the obtained guide image will damage the corners of the main structure, but the smoothing effect of the obtained image is better; and when only transform domain filtering is used, the obtained guide image can be better The corners of the main structure are preserved, but the resulting image is less smoothed. Therefore, this application uses both bilateral filtering and transform domain filtering to process images, and the specific process can be shown as follows:

S2011: Obtain the original image, the constant value image, the spatial weight value and the range weight value;

S2012: Using the constant value image as the initial guiding function, perform transform domain filtering on the original image and the initial guiding function according to the transform domain filtering method to obtain a new guiding function;

S2013: Perform bilateral filtering on the original image and the new guiding function according to the bilateral filtering method to obtain a primary guiding function;

S2014: Repeat S2012 and S2013 for the preset number of times to obtain a guide image.

The original image is the image to be smoothed, and can be an image in any format, such as tiff, tif, bmp, gif, etc., which will not affect the realization of the technical solution of this application.

A constant image is an image that is all constant 0s.

The spatial weight σ _s and the range weight σ _r are the parameters used when filtering the image. Generally, when using transform domain filtering, the spatial weight can be set to 1.5 times the spatial weight of the bilateral filter, and the range weight can be set to the bilateral filter. 3 times the space weight. Of course, it is also possible not to set the parameters according to the above, and those skilled in the art can configure the parameters according to the actual situation, which will not affect the implementation of the present application.

Regarding the value of the spatial weight and range weight (spatial weight σ _s > 1, range weight σ _r > 0), the impact on image smoothing and resolution, please refer to Figure 3-5, Figure 3 is the original image to be processed image. It can be seen from Figure 4-5 that when other parameters remain unchanged, the image becomes more and more blurred as σ _s decreases; when other parameters remain unchanged, as σ _r increases, the image becomes more and more blurred. It can be seen from the figure that optionally, when the spatial weight value σ _s =3 and the range weight value σ _r =0.01, the image is clearer and has a better smoothing effect. Certainly, those skilled in the art may decide according to specific requirements and image resolution, which is not limited in this application.

It should be noted that bilateral filtering can be performed first, then transform domain filtering, and then looped n times; it can also be transformed first, then bilateral filtering, and then loop n times; of course, bilateral filtering can also be used for n The transformation domain filtering is used to perform n times of filtering; or the transformation domain filtering is used to perform n times of filtering, and then bilateral filtering is used to perform n times of filtering, which will not affect the implementation of this application. However, after many experiments, it is found that the smoothing effect of the obtained image is the best when it is first processed by transform domain filtering, then by bilateral filtering, and then looped n times. Therefore, optionally, transform domain filtering may be performed on the original image first, and then bilateral filtering may be performed, and then a preset number of cycles may be performed.

A complete filtering is to perform transformation domain filtering on the original image first, and then perform bilateral filtering. Loop execution refers to performing multiple operations on a complete filtering. For example, the guiding image obtained by a complete filtering is a guiding function. Perform another complete filtering to obtain the secondary guidance function; loop n times, and obtain the n-time guidance function.

According to the analysis of multiple experiments, there is only a slight change in the preset number of cycles exceeding 3 times, and increasing the number of times has no effect on the experimental results, but will increase the time of image processing, resulting in low image processing efficiency. Therefore, optionally, the preset number of times may be 3 times.

After performing the above operations in a loop, the details and small-scale noise of the original image can be blurred while retaining the structural components of the original image.

S202: Construct a least squares model for the original image, the guide image, and a preset smoothed image according to a least squares method.

The least square method (also known as the least square method) is a mathematical optimization method. It finds the best function matching of the data by minimizing the sum of the squares of the errors, and the unknown data can be easily obtained by using the least square method, and the sum of the squares of the errors between the obtained data and the actual data is minimized.

The smoothed image is an image obtained after the original image is smoothed by the image smoothing method of the present application. Because the smooth image is unknown, and the model constructed by the least square method can solve the unknown data, it can be assumed that the smooth image is a preset smooth image, which is brought into the least square model as unknown data. For example, the concept of a linear equation in one variable in mathematics, to solve the unknown quantity, the variable x is generally assumed first, and then the known quantity is brought into the solution to obtain the value of x.

A norm, is a function with the notion of "length" that assigns a non-zero positive length or magnitude to all vectors in a vector space. The semi-norm can instead assign zero length to non-zero vectors. For example, the Euclidean norm can be defined in the two-dimensional Euclidean geometric space R. Elements in this vector space are often drawn in a Cartesian coordinate system as directed line segments with arrows starting from the origin. The Euclidean norm of each vector is the length of the directed segment.

Norms can generally be divided into vector norms and matrix norms. In vector norms, 0-norm refers to the number of non-zero elements in the vector; 1-norm is the sum of the absolute values of vector elements; 2-norm The square root of the sum of the absolute values of the elements of a numeric vector. In the matrix norm, the 0-norm refers to the number of non-zero elements in the matrix; the 1-norm is the maximum value of the sum of the absolute values of all matrix column vectors; the 2-norm, also called the spectral norm, is Refers to the square root of the largest eigenvalue of the matrix, which is the modulus in the usual sense.

According to the least square method and the square of the 2-norm, the least square model can be constructed for the original image, the guide image and the preset smooth image as:

In the formula, S is the preset smooth image, I is the original image, G is the guide image, and α is the detail restoration factor.

It should be noted that α usually takes the value of α∈[0,1].

The constructed least squares model has a double data fidelity item, and one item is used to control the difference between the smoothed image and the original image (ie ), the other is used to control the difference between the smoothed image and the guided image Through these two controls, it can not only enhance the protection of the structural components in the original image, but also provide a detail restoration factor to restore some high-contrast details.

It should be noted that, compared with the use of one data fidelity item in the prior art (that is, used to control the difference between the smooth image and the original image), the two items of the double data fidelity item used in this application both retain the original image. The main structure, the difference is that one preserves details (first item) and the other blurs and weakens details (second item). The use of double data fidelity items can more effectively preserve the main structure, and at the same time preserve the high-contrast detail components to varying degrees according to the selection of different restoration parameters. If only the first data fidelity term is used, high-contrast details cannot be effectively filtered out, and the main structure is easily damaged. If only the second data fidelity term is used, all small components (which may contain useful components ).

S203: Using a norm to define the pixel intensity and gradient of the preset smooth image, so as to obtain a constraint function of the preset smooth image.

The 0-norm can be used to define the pixel intensity and gradient of the preset smooth image. Of course, the 1-norm and 2-norm can also be used for definition, which will not affect the implementation of this application. However, after many experiments and analysis, compared with 1-norm and 2-norm, 0-norm can achieve better results.

The 0-norm refers to the number of variables that are not zero, S is the preset smooth image, ▽S is the gradient map of the preset smooth image, and the 0-norm is used to define the pixel intensity and gradient of the preset smooth image is ||▽S|| ₀ , where, ▽S∈RM ^×N . Indicates the non-zero number in ▽S.

By constructing a constraint function, it can be used to control the sparsity of smooth images.

S204: Obtain a smooth energy objective function according to the least squares model and the constraint function.

According to the least squares model and the constraint function, the smooth energy objective function is:

In the formula, E(S) is the smooth energy objective function, S is the preset smooth image, I is the original image, G is the guide image, α is the detail restoration factor, λ is the smoothing factor, ||▽S|| is the constraint function .

When the constraint function is defined using the 0-norm, the above smooth energy objective function is:

where E(S) is the smooth energy objective function, S is the preset smooth image, I is the original image, G is the guide image, α is the detail restoration factor, λ is the smoothing factor, ||▽S|| ₀ is the constraint function.

Regarding the effect of the detail restoration factor and the smoothing factor (α∈[0,1], λ≥0) on the smoothing effect and resolution of the image, please refer to Fig. 6 and Fig. 7 . It can be seen from the figure that when the other parameters are constant, the image becomes more and more blurred with the decrease of α; when the other parameters are constant, the image becomes more and more blurred with the increase of λ. It can be seen from the figure that optionally, when α=1 and λ=0.005, the image is clearer and has a better smoothing effect. Certainly, those skilled in the art may decide according to specific requirements and image resolution, which is not limited in this application.

S205: Using the semi-quadratic splitting method and the alternating fixed variable method to solve the smooth energy objective function to obtain a smooth image.

Specific may include:

introducing auxiliary variables to replace the constraint function term in the smooth energy objective function;

Using the semi-quadratic splitting method to minimize the replaced smooth energy objective function, adding an error penalty term to obtain a smooth minimization model;

Solving the smoothing minimization model according to the alternating fixed variable method to obtain the smoothed image.

Since it is impossible to directly minimize the smooth energy objective function, which is a non-convex optimization problem, it is necessary to introduce auxiliary variables to replace the constraint function to make it as close to the minimum as possible.

For example, when the smoothed energy objective function is:

Introduce auxiliary variable g=(g _x , g _y ) ^T to replace ||▽S|| ₀ in the constraint item;

The replaced smooth energy objective function is minimized using the semi-quadratic splitting method, and the original smooth image function is:

Add an error penalty term to the above smooth image function to form the final smooth minimization model:

where β is an adaptive parameter to control the similarity between g and ▽S.

Alternate fixed variable method generally fixes one quantity and solves another quantity, which is an iterative process. That is, it is necessary to solve g and S alternately, and finally obtain a smooth image S. The specific process can be as follows:

Initialize the preset smooth image, the adaptive parameter β, and the number of iterations i;

Iterative calculations are performed using the following calculation relations:

β = kβ;

Until β>β _max , output S, which is the finally obtained smooth image.

Wherein, the preset smooth image is initialized as the original image, the adaptive parameter β is initialized as β ₀ , and k is the increase rate.

F ^-1 (·) represents the inverse discrete Fourier transform operator, F(·) represents the complex conjugate operator, and F(1) represents the discrete Fourier transform of the delta function. All of the above operators, addition, multiplication, and division, operate element-wise. Through the Fourier transform, the speed of solving S is accelerated, which is beneficial to improve the efficiency of the overall image smoothing process.

Optionally, β ₀ =λ, β _max =10 ⁵ , k=2. Certainly, those skilled in the art may decide according to specific requirements and image resolution, which is not limited in this application.

It can be seen comprehensively that in the image smoothing processing method provided by the present application, S201 is local smoothing, and S202-S205 are global smoothing. The local smoothing method can effectively blur small high-contrast components and retain large structures; the global smoothing method can effectively remove small structures blurred by local filtering and retain large structures, eventually resulting in the retention of large results in the original image , small detail noise removal.

It can be seen from the above that the embodiment of the present invention comprehensively considers the global features and local features, first performs local smoothing processing on the original image, and then performs global smoothing processing, avoids the disadvantages of the local smoothing method and the global smoothing method, and effectively utilizes the advantages of both Advantage. By controlling the difference between the smoothed image and the original image and controlling the difference between the smoothed image and the guided image, the protection of the structural components in the original image is enhanced, the structure of the image is preserved, and some high-contrast details are restored while removing detailed texture features. , to obtain a good image smoothing effect; in addition, effective noise filtering is carried out on the image, which strengthens the intensity of boundary pixels, which is beneficial to the extraction of image contours, and thus helps to improve the accuracy and efficiency of image recognition.

When the image contains a lot of high-contrast noise, when constructing the least squares model, because the original function is too noisy, the constructed model is too much disturbed by the noise, and the smooth image obtained by solving will have a large deviation, and it is difficult to ensure smoothness The accuracy of the post image. Therefore, the present application also provides an embodiment based on the above embodiment.

Before constructing the least squares model, median filtering is performed on the original image.

The median filtering method is a nonlinear smoothing technique, a nonlinear signal processing technique that can effectively suppress noise based on the sorting statistics theory. The median value of all pixel gray values. The basic principle of median filtering is to replace the value of a point in a digital image or digital sequence with the median value of each point in a neighborhood of the point, so that the surrounding pixel values are close to the true value, thereby eliminating isolated noise points. The method is to use a two-dimensional sliding template of a certain structure to sort the pixels in the board according to the size of the pixel value, and generate a monotonically rising (or falling) two-dimensional data sequence.

That is, the least squares model is constructed on the original image after median filtering, the guide image and the preset smoothing image according to the least square method. Specifically, the implementation method is the same as that of the foregoing embodiment, and will not be repeated here.

After using the median filter to process the original image, it can effectively filter out the high-contrast noise in the original image, improve the smoothing effect of the image, and improve the accuracy of the smoothed image.

The embodiment of the present invention also provides a corresponding implementation device for the image smoothing method, which further makes the method more practical. The image smoothing device provided by the embodiment of the present invention is introduced below, and the image smoothing device described below and the image smoothing method described above can be referred to in correspondence.

Referring to FIG. 8, FIG. 8 is a structural diagram of an image smoothing device provided by an embodiment of the present invention, which may include:

The filtering and smoothing module 801 is configured to circularly perform bilateral filtering and transform domain filtering for a preset number of times on the original image to obtain a guide image.

The model building module 802 is configured to construct a least square model for the original image, the guide image and the preset smooth image according to the least square method.

Obtaining a smooth image module 803, configured to define the pixel intensity and gradient of the preset smooth image using a norm, so as to obtain a constraint function of the preset smooth image; smoothing is obtained according to the least squares model and the constraint function Energy objective function: solving the smooth energy objective function by using semi-quadratic splitting method and alternating fixed variable method to obtain a smooth image.

In a specific implementation manner, the module 803 of obtaining a smoothed image is to construct a least squares model for the original image, the guide image and the preset smoothed image according to the least square method and the square of the 2-norm module, the least squares model is:

In the formula, S is the preset smooth image, I is the original image, G is the guide image, and α is the detail restoration factor.

The functions of each functional module of the image smoothing device in the embodiment of the present invention can be implemented according to the method in the above method embodiment, and the specific implementation process can refer to the relevant description of the above method embodiment, and will not be repeated here.

It can be seen from the above that the embodiment of the present invention comprehensively considers the global features and local features, first performs local smoothing processing on the original image, and then performs global smoothing processing, avoids the disadvantages of the local smoothing method and the global smoothing method, and effectively utilizes the advantages of both Advantage. By controlling the difference between the smoothed image and the original image and controlling the difference between the smoothed image and the guided image, the protection of the structural components in the original image is enhanced, the structure of the image is preserved, and some high-contrast details are restored while removing detailed texture features. , to obtain a good image smoothing effect; in addition, effective noise filtering is carried out on the image, which strengthens the intensity of boundary pixels, which is beneficial to the extraction of image contours, and thus helps to improve the accuracy and efficiency of image recognition.

In order to verify that the technical solution provided by this application has a good image smoothing effect, this application provides specific examples, please refer to Figures 9-14, Figure 9 is the original image to be processed, and Figures 10-13 are processed by other algorithms Image, Figure 14 is the image processed in this application, and the picture in the box in the figure is a partial enlarged picture of the corresponding box. It can be seen from the figure that in the image processed by the RTV (Relative Total Variation, correlation total variation) algorithm, the two similar lines are blurred and cannot be recognized; other algorithms (such as RGF (Rolling Guidance Filter, circular guidance filter) and BLF ( Bilateral filter, bilateral filtering) algorithm, NLGRTV (Nonlocal version of Generalized Relative Total Variation, non-local version of the general correlation total variation) algorithm and SSPTF (Scale-aware Structure-Preserving Texture Filtering, scale-aware structure-preserving texture filtering) algorithm) The smoothing effect on image detail is coarser. It can be seen that the method of the present application has a good image smoothing effect.

In order to verify that the technical solution provided by this application has an effective denoising effect, this application provides specific examples, please refer to Figures 15-17, Figure 15 is the original image to be processed, and Figure 16 is the image processed by the RGF and BLF algorithms Figure 17 is the image processed by this application. It can be seen from the figure that there are still blurred and messy lines in Figure 16. The method provided by this application effectively removes the messy lines in the original image and obtains the target object (cuboid structure) object). It can be seen that the technical solution provided by the present application can effectively filter image noise and avoid the interference of image noise.

In order to verify that the technical solution provided by this application has the effect of enhancing images, this application provides specific examples, please refer to Figures 18 and 19, Figure 18 is the original image, Figure 19 is the image processed by this application, as can be seen from the figure, after In the method provided by the present application, the pixels of the original image are significantly enhanced, especially the intensity of edge pixels, which is beneficial to extracting the contour of the image, thereby improving the accuracy and efficiency of image recognition.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related part, please refer to the description of the method part.

Professionals can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the possible For interchangeability, in the above description, the composition and steps of each example have been generally described according to their functions. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present invention.

The steps of the methods or algorithms described in connection with the embodiments disclosed herein may be directly implemented by hardware, software modules executed by a processor, or a combination of both. Software modules can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other Any other known storage medium.

The image smoothing method and device provided by the present invention have been introduced in detail above. In this paper, specific examples are used to illustrate the principle and implementation of the present invention, and the descriptions of the above embodiments are only used to help understand the method and core idea of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, some improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims (10)

1. a kind of image smoothing method, it is characterised in that including：

Bilateral filtering and the transform domain filtering of preset times are carried out to original image circulation, to obtain navigational figure；

Least square mould is constructed to the original image, the navigational figure and default smoothed image according to least square method Type；

The image pixel intensities and gradient of the default smoothed image are defined using norm, to obtain the pact of the default smoothed image Beam function；

Smoothed energy object function is obtained according to the least square model and the constraint function；

The smoothed energy object function is solved using half secondary split method and alternating fixed variable method, to obtain smooth figure Picture.

2. image smoothing method according to claim 1, it is characterised in that it is described according to least square method to described original Image, the navigational figure and default smoothed image construct least square model：

According to the least square method and 2- norms square to the original image, the navigational figure and default smooth Image configuration least square model is：

$\mathrm{\α}\left|\right|S-I|{|}_2^2+(1-\mathrm{\α})||S-G|{|}_2^2;$

In formula, S is the default smoothed image, and I is the original image, and G is the navigational figure, and α is the detail recovery factor.

3. image smoothing method according to claim 1, it is characterised in that the utilization norm defines described default smooth The image pixel intensities and gradient of image are：

The image pixel intensities and gradient of the default smoothed image are defined using 0- norms.

4. image smoothing method according to claim 3, it is characterised in that it is described according to the least square model and The constraint function obtains smoothed energy object function：

Obtaining smoothed energy object function according to the least square model and the constraint function is：

$E\left(S\right)=\mathrm{\α}\left|\right|S-I|{|}_2^2+(1-\mathrm{\α})||S-G|{|}_2^2+\mathrm{\λ}\left|\right|\▿S|{|}_0;$

In formula, E (S) is the smoothed energy object function, and S is the default smoothed image, and I is the original image, and G is institute Navigational figure is stated, α is the detail recovery factor, and λ is smoothing factor, | | ▽ S | |₀It is the constraint function.

5. the image smoothing method according to Claims 1-4 any one, it is characterised in that described according to least square Method constructs least square model to the original image, the navigational figure and default smoothed image：

According to the least square method to the original image by medium filtering, the navigational figure and default smoothed image structure Make least square model.

6. image smoothing method according to claim 5, it is characterised in that described that default following is carried out to original image circulation Bilateral filtering and the transform domain filtering of ring number of times, are included with obtaining navigational figure：

S1：Obtain the original image, constant value image, space weighted value and scope weighted value；

S2：Be initial guide function with the constant value image, according to transform domain filter method to the original image and it is described just Beginning guidance function carries out transform domain filtering, obtains new guidance function；

S3：Bilateral filtering is carried out to the original image and the new guidance function according to bilateral filtering method, is once drawn Derived function；

S4：The preset times are performed to S2 and S3 circulations, to obtain the navigational figure.

7. image smoothing method according to claim 6, it is characterised in that the preset times are 3 times.

8. image smoothing method according to claim 7, it is characterised in that the space weighted value and scope weighted value For：

The space weighted value is 3；

The scope weighted value is 0.01.

9. image smoothing method according to claim 8, it is characterised in that described to utilize half secondary split method and alternating Fixed variable method solves the smoothed energy object function, is included with obtaining smoothed image：

It is introduced into the constraint function that auxiliary variable is replaced in the smoothed energy object function；

Minimum treatment is carried out to the smoothed energy object function replaced using the half secondary split method, error punishment is added , obtain smooth minimum model；

The smooth minimum model is solved according to the alternately fixed variable method, to obtain the smoothed image.

10. a kind of image smoothing device, it is characterised in that including：

Filtering module, bilateral filtering and the transform domain filtering for carrying out preset times to original image circulation, to obtain Obtain navigational figure；

Model module is set up, for scheming to the original image, the navigational figure and default smoothing according to least square method As construction least square model；

Smoothed image module is obtained, image pixel intensities and gradient for defining the default smoothed image using norm, to obtain Obtain the constraint function of the default smoothed image；Smoothed energy is obtained according to the least square model and the constraint function Object function；The smoothed energy object function is solved using half secondary split method and alternately fixed variable method, it is flat to obtain Sliding image.