CN106203511B - A kind of image similar block appraisal procedure - Google Patents

Abstract

A method for evaluating image similarity blocks, comprising the following steps: (1) taking two image blocks S _A and S _B whose size is K*K in a cryo-electron microscope image; (2) calculating each pixel point in the image block S _A The corresponding gradient value tAi is stored in the matrix DS _A ; the gradient value tBi corresponding to each pixel in the image block S _B is calculated and stored in the matrix DS _B ; (3) the weight weight1 of the i-th point is calculated, weight2; (4) Calculate the geodesic distance between Ai and Bi at the same position in the image block; (5) Calculate the geodesic distance between S _A and S _B ; (6) Compare d( _{SA, S B )} Determine whether the image blocks S _A , S _B are similar to the size of the predefined threshold T. The method of measuring similar blocks based on geodesic distance can effectively improve the accuracy of similar block groups, and it can be used in the denoising algorithm based on image blocks to improve the denoising effect.

Description

An Image Similarity Block Evaluation Method

technical field

The invention relates to the field of computers, in particular to a method for evaluating similar blocks of cryo-electron microscope images.

Background technique

With the rapid development of science and technology, people's understanding of biology is no longer limited to the macroscopic stage. People want to have a deeper understanding of the microscopic world of biology. According to the structure to determine the function, it is necessary to understand the function and internal mechanism of biological macromolecules, such as To copy and reproduce, we must first know its structure. Due to the complexity of the internal structure and huge molecular weight of biological samples, analyzing its three-dimensional structure has always been the bottleneck that affects the follow-up research on its function. Until the middle and late last century, with the rapid development of material science, more and more precision instruments and equipment gradually The emergence of precision instruments and equipment such as cryo-electron microscopes is gradually used in the study of biological samples.

After decades of development, the techniques and methods for determining the structure of biological macromolecules have been greatly developed. There are mainly three types: X-ray crystallography, NMR (nuclear magnetic resonance spectroscopy), and cryo-electron microscopy (referred to as Cryo-electron microscopy, Cryo-EM technology), X-ray crystallography requires the crystallization of biological samples to be tested, and NMR requires the purification of biological samples. Both methods have limitations because of the complexity and molecular weight of biological samples. It is huge, and analyzing its three-dimensional structure has always been the bottleneck affecting its subsequent functional research. Until the middle and late last century, with the rapid development of material science, more and more precision instruments and equipment gradually appeared, and electron microscopes were gradually used in biological research. Sample study.

One disadvantage of cryo-EM imaging is that many biological samples are sensitive to radiation, and the electron beam may damage the sample. To prevent this from happening, the sample is usually imaged under a very low electron beam. However, this tends to cause image The signal-to-noise ratio is very low, the noise of the image is enhanced, and the obtained image is often polluted by Poisson noise and Gaussian noise. In addition, environmental factors can also cause image noise, such as magnetic field changes, mechanical vibration, acoustic vibration, and thermal noise. Stability, electromagnetic lens, etc. In addition, the noise is also related to the galvanometer, which approximates the Poisson distribution, and Gaussian noise will also be brought in when the analog projection image is converted into a digital image. noise. In addition, the noise in the image will also affect the single particle selection and calibration steps in the three-dimensional reconstruction of the cryo-electron microscope image. Therefore, the main purpose of denoising the cryo-electron microscope image is to preserve as much detail information of the image as possible while reducing the noise. to improve image quality. Commonly used electron microscope image denoising methods, such as Gaussian filter technology, can eliminate noise, but this method also makes the edges of the image blurred, which reduces the image quality as a whole, and the image edges are not clear enough, which seriously affects the subsequent particles. pick.

To obtain the structure of biological macromolecules, we first need to obtain their projection images. However, cryo-electron microscopy images taken by cryo-electron microscopy have the characteristics of weak phase, low signal-to-noise ratio, low contrast, uneven background intensity, and irregular texture inside particles. , which brings great difficulties to accurately determine the three-dimensional structure of biological macromolecules. In order to solve this problem, it is necessary to denoise the projection image, improve the visual effect of the image, and restore the image quality to the maximum extent. On the basis of analyzing the original electron cryo-EM images, we found that the denoising method based on the image block can effectively eliminate the noise components in the original image, but at the same time, we also found that the method first needs to find the image block similar to the reference block, while How to evaluate whether two image blocks are similar becomes the key to our solution to this problem. In many classic image processing algorithms based on image blocks, Euler distance is often used to evaluate whether two image blocks are similar, but this method has certain limitations, because Euler distance only considers the grayscale of the image. At the same time, the image block subspace is not completely Euler space, and using Euler distance is not a good measure of similarity. Therefore, when selecting similar blocks, it is necessary to consider the structure of the image subspace and some structures feature.

Therefore, it is imperative to find a simple and efficient method that can accurately evaluate similar blocks.

Contents of the invention

Image block matching (Block-Matching) algorithm is based on the redundancy and correlation of image information. By calculating the distance between the candidate block set X and the reference block set R, the reference block class to which the candidate block belongs is found. The block-matching algorithm in the market has the K nearest neighbor search method, which belongs to the local search, and the search speed is relatively fast, but it can only obtain the local optimal solution. Another matching method is the global search method, although the global optimal solution can be obtained, but This method takes a long time, and the quality of the matching algorithm has seriously affected other subsequent processing. The image block subspace is not completely Euler space. When judging similar blocks, it is necessary to consider the structure of the image subspace and the structural characteristics of the image block.

The invention proposes an image similar block evaluation method, which is a similar block matching algorithm based on geodesic distance, and uses geodesic distance instead of Euler distance to judge whether two image blocks are similar. The method for measuring similar blocks based on the geodesic distance proposed by the invention is used in a denoising algorithm based on image blocks, which can effectively improve the accuracy of similar blocks and improve the denoising effect.

A method for evaluating image similarity blocks, comprising the following steps:

(1) Take two image blocks S _A and S _B whose size is K*K in the cryo-electron microscope image;

(2) Calculate the gradient value tAi corresponding to each pixel in the image block S _A , and save it in the matrix DS _A ; calculate the gradient value tBi corresponding to each pixel in the image block S _B , and save it in the matrix DS _B middle;

(3) Calculate the weight weight1 and weight2 of the i-th point:

weight1＝0.5*(value[Ai]-value[Bi]) ² where: value[Ai] and value[Bi] respectively represent the gray values of Ai and Bi corresponding to the same position in image blocks S _A and S _B ;

weight2=0.5*(tAi+tBi+tanα-β) where: tAi, tBi respectively represent the gradient value of the i-th pixel in the image blocks S _A and S _B ; α, β represent the change of the i-th pixel pixel value respectively The angle between the maximum direction and the minimum direction of change;

(4) Calculate Ai at the same position in the image block, the geodesic distance between two points Bi:

d(S _Ai ,S _Bi )=weight1+weight2;

(5) Calculate the geodesic distance between S _A and S _B :

Where: i is the i-th pixel in the image block, i=1,2...K*K;

(6) Compare the size of d( _{SA, S B )} with the predefined threshold T, and judge whether the image blocks _{SA, S B are} similar.

In the present invention, the two image blocks described in step (1) are randomly selected from the whole cryo-EM image.

In the present invention, the K represents the size of the image block. There are K*K pixels in the image block with a size of K. K is 2-50, preferably 5-20, more preferably 6-10.

In the present invention, the gradient value in step (2) is obtained by the gradient function in matlab.

In the present invention, α described in step (3) is [0, π].

In the present invention, β is [0, π].

In the present invention, the grayscale value described in the step (3) is obtained by the following method: the cryo-electron microscope image itself is a grayscale image, and by analyzing the header file format, after reading with matlab, the numerical value in the matrix is exactly the value of the image. grayscale value.

In the present invention, the value of K in step (5) is the same as the value of K in step (1).

In the present invention, the value of T in step (6) is related to the size K of the image block.

In the present invention, T is less than 2K*K, and T is 2-70, preferably 5-60, more preferably 10-50.

In the present invention, in step (6):

1) If d(S _A , S _B )≤T, the two image blocks are similar;

2) If d(S _A , S _B )>T, the two image blocks are not similar.

In the present invention, two image blocks S _A and S _B of size K*K are randomly selected from a cryo-EM image of size N*N, and the above method is used to judge whether the two image blocks are similar.

In the present invention, the N is 2 ^m , and m is 4-50, preferably 5-20, more preferably 6-15.

In the present invention, N*N represents the image size of the cryo-electron microscope.

In the present invention, the similarity between two image blocks randomly selected in the image is measured by calculating the geodesic distance between them, and the processing object is the image block.

In the present invention, in addition to the gray value of the image, the calculation method of the geodesic distance also considers the influence of the gradient value of the image on the geodesic distance, and the specific calculation formula is a joint expression containing the gray value of the image and the gradient.

In the present invention, the calculation formula of the geodesic distance considers the influence of the pixel value and the gray value on the geodesic distance at the same time, each accounting for a 1/2 ratio. ,π] interval increasing sine function to describe the two-point gradient change.

In the present invention, what is calculated is the geodesic distance between each pixel point in the image block and the pixel point at the same position of another image block, and iterates multiple times until all the pixel points participate in the calculation, iterates but does not update Image blocks.

In the present invention, the geodesic distance between K*K points in the image block is calculated first, and then the geodesic distance between two image blocks is obtained by weighted average, and whether the image blocks are similar is judged by the geodesic distance, mainly It is used in the design of denoising algorithm based on image blocks.

In the present invention, block matching is the simplest and most effective method for finding similar blocks, but this method is inefficient. In actual image processing applications, a local window slightly larger than the reference block is usually used to replace the global window. The basic idea of finding similar blocks using a local sliding window method is to assume that many similar blocks of the reference block can be found in a relatively small area. However, in practice, some salient features of the image, such as corners, round edges , will not appear in a certain neighborhood, and using local similarity in non-repeating patterns may cause large errors. In this case, it is more appropriate to search for similar blocks globally in the entire image domain. In order to solve these two problems, the present invention proposes a new solution method, which uses geodesic distance instead of Euler distance to judge whether two image blocks are similar, and then gathers all similar blocks into similar block groups, Find the mean value of the similar block group, and then subtract the mean value of the similar block group from each image block. The DC component of the similar block can be removed through the subtraction operation, but the important structural characteristics of the image block will not be changed, and then all subtracted Image patches with DC components removed for prior learning.

In the present invention, in the existing similar block matching algorithm, the candidate block S _x and the reference block are usually used The Euler distance between them is used to measure their similarity. However, when there are a large number of points that are not in the calculation area between two points, it is usually invalid to use the Euler distance to calculate the distance between the two points, because it does not take into account the local connectivity, so the Euler distance has a space In order to overcome this limitation, in this invention, we will use the geodesic distance instead of the Euclidean distance, and measure the similarity by calculating the geodesic distance between two image blocks. The geodesic distance is used for data processing, usually in classification and similarity comparison, and the present invention measures the similarity by calculating the geodesic distance between two image blocks. In the image domain, an image can be represented by a two-dimensional discrete function. The gradient of the image is obtained by a two-dimensional discrete function. The direction of the gradient is the direction of the maximum change of the gray value. Therefore, the gradient value of two points can be used to describe the geodesic distance. α, β are the angles between the two gradient directions, and the value range is [0, π], the angle between the direction of the maximum gray scale change and the minimum gray scale change direction.

In the present invention, although the wavelet transform effectively utilizes the sparse attribute of the natural image, it depends on the wavelet basis function adopted during the wavelet decomposition, lacks translation invariance, and will lose a large amount of detailed information when removing noise. However, the existing block-based BM3D algorithm uses hard threshold and Wiener filtering. This method depends on the selection of the threshold, and in this method, the Euclidean distance is used to measure similar blocks, which has certain limitations. Based on the global dictionary Image denoising does not fully consider the prior knowledge of non-local self-similarity between image patches in natural images.

When the idea of prior learning is combined with the sparse and redundant representation of images, the denoising method based on prior learning of image similar blocks uses dictionaries to denoise images. The theoretical basis of dictionary denoising is that the ideal image has a sparse representation under an appropriate over-complete dictionary, and the noise destroys this sparse representation. By selecting or designing an appropriate dictionary, the sparse representation of the natural image under the dictionary can be obtained. To reduce or eliminate noise. One of the prior knowledge of using dictionary denoising is the sparsity of the signal. The denoising method based on the prior knowledge of image blocks can effectively retain the local information of the image to achieve better experimental results. The dictionary obtained through learning is better than using a fixed Sparse basis has better denoising performance.

There are a large number of identical particles in the MRC image. Blocking the image can effectively use the properties of these identical particles and achieve better experimental results. Therefore, the present invention combines the similar block matching method based on geodesic distance and the non-local self-similar prior knowledge (NSS) of the image block, and performs prior learning on the NSS, constructs a dictionary of similar block groups, and solves the weighted sparse coding model. The optimal solution is to obtain the sparse coding of similar block groups, use the sparse representation of the image to denoise the image blocks, and finally reconstruct all the image blocks to obtain the denoised image. This method fully considers the distance calculation method in the manifold space, uses the geodesic distance to accurately select similar blocks, and at the same time, takes into account the internal prior knowledge (NSS) and external prior knowledge (sparseness) between similar blocks.

The purpose of MRC image denoising is to eliminate noise in the image, improve image contrast and signal-to-noise ratio, and provide sufficient effective information for subsequent single particle selection and two-dimensional projection image classification. Assume to observe image y, noise-free image x, and noise v, then there is y=x+v, v～N(0,σ ² ), therefore, the problem of image denoising is transformed into obtaining the estimated value of image x by observing image y make The minimum, that is, the minimum mean square error MSE, so that the maximum PSNR value can be obtained to achieve the optimal denoising effect.

The non-local mean can effectively preserve the detailed information of the image while eliminating the noise, and can be used for many different images. In order to effectively utilize the structural information of the image and the properties of the non-local mean value that can effectively eliminate noise, the present invention combines the characteristics of the cryo-electron microscope image to group and block the image, extract several image blocks from an MRC image, and select N As a reference block, gather all similar blocks into similar block groups. Here, the geodesic distance mentioned above is used to measure similar blocks, and then all similar blocks are aggregated into a similar block group. There are a total of N similar block groups. Each A similar block group contains M similar blocks, and y _m represents the image block in image y, and x _m represents the image block in image x. Therefore, the image denoising problem can be transformed into the problem of finding the minimum MSE, that is

in Because the non-local mean can suppress the noise, the dictionary representation can effectively represent the non-noise signal in the image. Therefore, combining the non-local mean and the dictionary representation, we know Therefore by solving the dictionary D and the sparse coding coefficients The effect of denoising image blocks is achieved, and finally the MRC image after denoising is obtained by aggregating all image blocks.

Grouping images into blocks has the following advantages: it can make full use of the prior information between image blocks and model them, which is more suitable for parallel computing and improves efficiency. Fig. 4 is a flow chart of denoising an MRC image using the method proposed by the present invention.

use Represents M similar blocks of size p*p in the observed image y, where use Indicates the mean value of these M image blocks, The next step is to Perform prior learning to obtain K Gaussian distributions.

In image patch-based processing methods, it is generally considered that all image patches are independently sampled. Each GMM model consists of K Gaussian distributions, through all noise-free image blocks Carry out learning to obtain K Gaussian components. In the present invention, considering that the mean value operation on the image block can suppress noise and retain the structural characteristics of the image block to the greatest extent, the present invention adopts the method of The method of performing prior learning obtains K Gaussian components. Therefore, in the present invention The likelihood function of is expressed as Here π _k is the weight factor, indicating the probability that the kth Gaussian component is selected, In the present invention, N similar block groups are independent of each other, so the global target likelihood function can be expressed as For the convenience of calculation below, take the logarithm of the target likelihood function The usual approach is to take the derivative of the target likelihood function and solve the equation by setting its derivative to 0. However, the logarithmic function cannot be simplified for the accumulation operation containing K Gaussian components in the lnL expression, so the maximum value cannot be obtained directly by derivation. The present invention selects points randomly from the GMM model and solves the problem through EM algorithm.

Through GMM learning, K Gaussian distributions that can describe the structural characteristics of image blocks can be obtained. Each Gaussian distribution is called a Gaussian component. In the following, the Bayesian method is used to select the appropriate Gaussian component for each similar block group, and the similarity Block group denoising. For each group of similar blocks Computes the probability that it was generated by the kth Gaussian component. Because groups of similar blocks have the same Gaussian distribution. Assume that the kth Gaussian component can effectively describe According to the Bayesian probability formula, we have take the logarithm here

For different values of k, C is the same. For each value of k, k=1,2,...K, calculate the logarithmic value of the maximum posterior probability When comparing different values of k, size, choose to enable The k value for which the maximum value is obtained, the corresponding Gaussian component is selected for the Perform follow-up processing. At this time, Σ _k represents the covariance matrix of the kth Gaussian component, and performs singular value decomposition (SVD) on Σ _k , that is, Σ _k = DΛD ^T , where D is the orthogonal eigenvector matrix, and Λ is the eigenvalue Diagonal matrix, the eigenvector D characterizes the non-local self-similar statistical structure, therefore, D can be used as a sparse coding dictionary, and α is used to represent the sparse coding coefficient, so In the present invention, the constraints of the sparse coding model can be expressed as Use α to represent the coefficient of sparse coding, v is noise, w is the weight vector of α, Indicates a 2-norm, ||w ^T α|| ₁ indicates a 1-norm, D is an orthogonal matrix, ^DDT =I, |D|=±1, and I is an identity matrix. According to the maximum a posteriori probability MAP, we know that According to Bayes formula, Available The noise v satisfies the v～N(0,σ ² ) distribution, therefore, according to the Gaussian distribution probability density function, we have The sparse coding coefficient α satisfies the Laplace distribution, therefore, in c is a constant, so Available where ε is a positive number close to 0. Because D is an orthogonal matrix, ^DDT =I, |D|=±1, so Pick Available also because so make Therefore, the sparse coding constraints can also be expressed as Deriving α _i , we can get which is Usually written in the following form Where (a)+=max(a,0), sgn(z _i ) is a sign function. Define the function SoftMAP(g _i ,τ _i )=sgn(g _i )(|g _i |-τ _i ) ₊ , so so

The noise in the image block and the inaccurate similar block set will affect the GMM learning, which in turn affects the dictionary and the weighted sparse coding matrix. The geodesic distance proposed by the present invention can improve the accuracy of the similar block group, thereby improving the correct rate of the dictionary , and finally use the dictionary D and weighted sparse coding The combination of get denoised image block

In the present invention, first use the GMM model to Perform prior learning to get K Gaussian components, when the kth Gaussian component is selected as When the most suitable Gaussian component of , we decompose its covariance Σ _k by SVD to obtain the dictionary D, solve the corresponding weighted sparse coding according to the sparse coding model, and then denoise the image block groups respectively.

Use the method of the present invention to obtain the estimated value of each image block in the MRC image, and finally reconstruct all image blocks to obtain a denoised image When multiple estimates occur at a certain location in the image, the final estimate is calculated by weighted average. By updating the noise variance Multiple iterations are performed, and η is a constant, thereby improving the denoising effect.

Compared with the prior art, the technical solution of the present invention has the following technical effects:

1. The calculation is simple and fast. Since the image block is composed of a number of pixels, the geodesic distance between the image blocks is obtained by calculating the geodesic distance between points for cumulative evaluation. During the operation, the gradient function compiled in matlab is directly called, saving the overall calculating time.

2. Higher accuracy. Traditional evaluation methods, such as Euclidean distance, but because the image block subspace is not completely Euler space, using Euler distance is not a good measure of similarity, while geodesic distance based on manifold space is It can better describe the subspace, use the geodesic distance as the similarity measure, and search all similar blocks more accurately.

3. Euler space is a special case of manifold space. If it is an adjacent point in the original image, the Euclidean distance is used instead of the geodesic distance. Otherwise, the geodesic distance is calculated using the distance formula mentioned above.

4. The denoising algorithm of the technical solution of the present invention is essentially capable of removing noise to the greatest extent while maintaining the integrity of the effective information of the original image, and has relatively low computational time complexity and space complexity .

5. The technical solution of the present invention is based on the idea of image blocks and effectively utilizes the properties of identical particles in cryo-electron microscope images, and the geodesic distance takes into account both the gray value and gradient value of the image, and the expression is more accurate.

Description of drawings

Fig. 1 is a flowchart of the similarity evaluation of two image blocks in the present invention.

Fig. 2 is a schematic diagram of similar blocks in an embodiment of the present invention.

Fig. 3 is a schematic diagram of the operation of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation examples described here are only used to explain the present invention, and are not intended to limit the present invention.

A method for evaluating similarity of image blocks, comprising the following steps:

In step A, two image blocks of the same size are randomly selected in a cryo-EM image.

A1. Store the pixel values of the two image blocks in matrix respectively.

A2. Use matrix to store the gradient values of two image blocks respectively

Step B, calculate the geodesic distance:

B1, calculate the weight weight1, weight2

B2. Calculate the geodesic distance of the corresponding point in the image block;

B3. Calculate the geodesic distance between image blocks

Step C, compare the geodesic distance with the fixed threshold T:

Step D, judging whether two image blocks are similar:

More specifically:

A method for evaluating image similarity blocks, comprising the following steps:

(1) Take two image blocks S _A and S _B whose size is K*K in the cryo-electron microscope image;

(2) Calculate the gradient value tAi corresponding to each pixel in the image block S _A and store it in the matrix DS _A ; calculate the gradient value tBi corresponding to each pixel in the image block S _B and store it in the matrix DS _B ;

((3) Calculate the weight weight1, weight2 of the i-th point:

weight1＝0.5*(value[Ai]-value[Bi]) ² where: value[Ai] and value[Bi] respectively represent the gray values of Ai and Bi corresponding to the same position in image blocks S _A and S _B ;

weight2=0.5*(tAi+tBi+tan|α-β|), where: tAi, tBi represent the gradient value of the i-th pixel in image blocks S _A and S _B respectively; α, β represent the i-th pixel respectively The angle between the maximum direction and the minimum direction of pixel value change;

(4) Calculate Ai at the same position in the image block, the geodesic distance between two points Bi:

d(S _Ai ,S _Bi )=weight1+weight2;

(5) Calculate the geodesic distance between S _A and S _B :

Where: i is the i-th pixel in the image block, i=1,2...K*K;

(6) Compare the size of d( _{SA, S B )} with the predefined threshold T, and judge whether the image blocks _{SA, S B are} similar.

In the present invention, the two image blocks described in step (1) are randomly selected from the whole cryo-EM image.

In the present invention, the K represents the size of the image block. There are K*K pixels in the image block with a size of K. K is 2-50, preferably 5-20, more preferably 6-10.

In the present invention, the gradient value in step (2) is obtained by the gradient function in matlab.

In the present invention, α described in step (3) is [0, π].

In the present invention, β is [0, π].

In the present invention, the grayscale value described in the step (3) is obtained by the following method: the cryo-electron microscope image itself is a grayscale image, and by analyzing the header file format, after reading with matlab, the numerical value in the matrix is exactly the value of the image. grayscale value.

In the present invention, the value of K in step (5) is the same as the value of K in step (1).

In the present invention, the value of T in step (6) is related to the size K of the image block.

In the present invention, T is less than 2K*K, and T is 2-70, preferably 5-60, more preferably 10-50.

In the present invention, in step (6):

1) If d(S _A , S _B )≤T, the two image blocks are similar;

2) If d(S _A , S _B )>T, the two image blocks are not similar.

In the present invention, two image blocks S _A and S _B of size K*K are randomly selected from a cryo-EM image of size N*N, and the above method is used to judge whether the two image blocks are similar.

In the present invention, said N is 2 ^m , m is 4-50, preferably 5-20, more preferably 6-15.

In the present invention, N*N represents the image size of the cryo-electron microscope.

Example 1

As can be seen from Figure 3, assuming that the two included angles are α and β respectively, α represents the angle between the direction of the maximum change in the gray value of point Ai and the direction of the minimum change in gray value, and β represents the direction of the maximum change in the direction of gray value of point Bi and the direction of gray value The angle between the minimum direction of value change defines the geodesic distance d(S _Ai , S _Bi ) between two points. Define weights weight1, weight2:

weight1=0.5*(value[Ai]-value[Bi]) ² Wherein: value[Ai], value[Bi] respectively represent the gray values of Ai and Bi at the same position in the image block;

weight2=0.5*(tAi+tBi+tan|α-β|) where: tAi, tBi represent the gradient value of the i-th pixel in the image blocks S _A and S _B respectively; α, β represent the i-th pixel respectively The angle between the maximum direction of pixel value change and the minimum direction of change;

Calculate the geodesic distance between two points Ai and Bi at the same position in the image block:

d(S _Ai ,S _Bi )＝weight1+weight2

An image block is composed of several pixels, given two image blocks S _A , S _B , the size is p*p, the geodesic distance between image blocks S _A , S _B

Among them, i is the i-th pixel in the image block, i=1,2...p*p, compare d(S _A ,S _B ) with the fixed threshold T, if d(S _A , S _B )<T, the image blocks S _A and S _B are considered to be similar blocks, and Fig. 1 is a flow chart of using the geodesic distance proposed by the present invention to measure similar blocks.

With cryo-electron microscope image N=1280, K=6, T=50 is taken as an example to illustrate the working process of the present invention.

As shown in Figure 1, the flow chart of calculating the geodesic distance between image blocks:

Including the following steps:

(1) Two image blocks S _A and S _B of size K*K are randomly selected from the cryo-EM image of size N*N.

(2) Calculate the geodesic distance between S _A and S _B :

Call the gradient function in matlab to calculate the gradient values of S _A and S _B respectively, and store them in the matrices DS _A and DS _B respectively.

Calculate the distance d _i of two points located at the same position in the image block, and store it in the matrix R

Calculate the geodesic distance between S _A and S _B

d(S _A ,S _B )＝32.07

(3) Compare the size of d(S _A , S _B ) and T, where T is 50:

d(S _A ,S _B )＜T

(4) Judging whether they are similar:

Image blocks S _A and S _B are similar blocks.

Claims (32)

1. A method for image similarity block evaluation, comprising the following steps: (1) Take two image blocks S _A and S _B whose size is K*K in the cryo-electron microscope image; (2) Calculate the gradient value tAi corresponding to each pixel in the image block S _A and store it in the matrix DS _A ; calculate the gradient value tBi corresponding to each pixel in the image block S _B and store it in the matrix DS _B ; (3) Calculate the weight weight1 and weight2 of the i-th point: weight1＝0.5*(value[Ai]-value[Bi]) ² where: value[Ai] and value[Bi] respectively represent the gray values of Ai and Bi corresponding to the same position in image blocks S _A and S _B ; weight2=0.5*(tAi+tBi+tan|α-β|) where: tAi, tBi represent the gradient value of the i-th pixel in the image blocks S _A and S _B respectively; α, β represent the i-th pixel respectively The angle between the maximum direction of pixel value change and the minimum direction of change; (4) Calculate Ai at the same position in the image block, the geodesic distance between two points Bi: d(S _Ai ,S _Bi )=weight1+weight2; (5) Calculate the geodesic distance between S _A and S _B : Where: i is the i-th pixel in the image block, i=1,2...K*K; (6) Compare the size of d( _{SA, S B )} with the predefined threshold T, and judge whether the image blocks _{SA, S B are} similar. 2. The method according to claim 1, characterized in that: the two image blocks described in step (1) are randomly selected in the whole cryo-electron microscope image, and/or The K represents the size of the image block, there are K*K pixels in the image block with the size K, and K is 2-50. 3. The method according to claim 2, characterized in that: K is 5-20. 4. The method according to claim 3, characterized in that: K is 6-10. 5. according to the method described in any one in claim 1-4, it is characterized in that: gradient value obtains by the gradient function in the matlab in the step (2). 6. The method according to any one of claims 1-4, characterized in that: α described in step (3) is [0, π], and/or β is [0, π]. 7. The method according to claim 5, characterized in that: α described in step (3) is [0, π], and/or β is [0, π]. 8. according to the method described in any one in claim 1-4,7, it is characterized in that: the grayscale value described in step (3) obtains by following method: cryo-electron microscope image itself is grayscale image, by analyzing The header file format, after being read by matlab, the value stored in the matrix is the gray value of the image. 9. The method according to claim 5, characterized in that: the grayscale value described in the step (3) is obtained by the following method: the cryo-electron microscope image itself is a grayscale image, and is read with matlab by analyzing the header file format Finally, the value stored in the matrix is the gray value of the image. 10. The method according to claim 6, characterized in that: the grayscale value described in step (3) is obtained by the following method: the cryo-electron microscope image itself is a grayscale image, and is read with matlab by analyzing the header file format Finally, the value stored in the matrix is the gray value of the image. 11. The method according to any one of claims 1-4, 7, 9-10, characterized in that the value of K in step (5) is the same as the value of K in step (1). 12. The method according to claim 5, characterized in that: the value of K in step (5) is the same as the value of K in step (1). 13. The method according to claim 6, characterized in that: the value of K in step (5) is the same as the value of K in step (1). 14. The method according to claim 8, characterized in that: the value of K in step (5) is the same as the value of K in step (1). 15. The method according to any one of claims 1-4, 7, 9-10, 12-14, characterized in that the value of T in step (6) is related to the image block size K. 16. The method according to claim 5, characterized in that: the value of T in step (6) is related to the image block size K. 17. The method according to claim 6, characterized in that: the value of T in step (6) is related to the image block size K. 18. The method according to claim 8, characterized in that: the value of T in step (6) is related to the image block size K. 19. The method according to claim 15, characterized in that: in step (6), T is less than 2K*K. 20. The method according to any one of claims 16-18, characterized in that: in step (6), T is less than 2K*K. 21. The method according to claim 19, characterized in that: in step (6), T is 2-70. 22. The method according to claim 10, characterized in that: in step (6), T is 2-70. 23. The method according to claim 21 or 22, characterized in that: in step (6), T is 5-60. 24. The method according to claim 23, characterized in that: in step (6), T is 10-50. 25. The method according to any one of claims 1-4, 7, 9-10, 12-14, 16-19, 21-22, 24, characterized in that: in step (6): 1) If d(S _A , S _B )≤T, the two image blocks are similar; 2) If d(S _A , S _B )>T, the two image blocks are not similar. 26. The method according to claim 5, characterized in that: in step (6): 1) If d(S _A , S _B )≤T, the two image blocks are similar; 2) If d(S _A , S _B )>T, the two image blocks are not similar. 27. The method according to claim 6, characterized in that: in step (6): 1) If d(S _A , S _B )≤T, the two image blocks are similar; 2) If d(S _A , S _B )>T, the two image blocks are not similar. 28. The method according to claim 8, characterized in that: in step (6): 1) If d(S _A , S _B )≤T, the two image blocks are similar; 2) If d(S _A , S _B )>T, the two image blocks are not similar. 29. The method according to claim 1, characterized in that: the size of the cryo-electron microscope image is N*N. 30. The method according to claim 29, characterized in that: said N is 2 ^m , m is 4-50, and/or N*N represents the size of the cryo-electron microscope image. 31. The method according to claim 30, characterized in that m is 5-20. 32. The method according to claim 31, characterized in that m is 6-15.