A fusion quality evaluation method based on q-order orthogonal fuzzy saliency

Abstract

The present application relates to a kind of fusion quality evaluation methods based on q order orthogonal fuzzy saliency, belong to image processing field, method includes: medical image is carried out block processing and obtains image subblock, again maximum pooling operation is carried out to image subblock;Image subblock after maximum pooling is carried out q order orthogonal fuzzy, and the q order orthogonal fuzzy closeness of image subblock is calculated again;The q order orthogonal fuzzy closeness of image subblock is combined with evaluation index, and the preference score of image subblock is obtained;According to image subblock priority ranking, the saliency coefficient of image subblock is obtained, and the closeness of image subblock is calculated;Based on the size of closeness, medical image is finally selected.The present application realizes the multi-dimension, accurate, robust evaluation of fusion quality, adapts infrared / visible light fusion, medical multi-modal fusion, remote sensing multispectral fusion and other multiple scenes, provides reliable support for fusion algorithm optimization and practical application, with wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more particularly to a fusion quality assessment method based on the saliency of q-order orthogonal fuzziness. Background Technology

[0002] Multimodal target tracking integrates image information from multiple medical sources using image fusion technology to achieve highly accurate and robust target tracking and detection. High-quality fusion results are a crucial prerequisite for multimodal target tracking, and fusion quality evaluation is an important method for screening high-quality fusion results, playing a vital role in ensuring the reliability and practicality of target tracking.

[0003] Currently, most mainstream evaluation methods are based on global calculations such as pixel statistics and structural similarity. These methods have limited ability to express fuzzy and uncertain information in images and rely too much on a single evaluation index, which can easily lead to discrepancies between the evaluation results and actual visual perception, making it difficult to provide effective evaluation information for multimodal target tracking. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a fusion quality assessment method based on q-order orthogonal fuzzy saliency, thus solving the deficiencies of the prior art.

[0005] The objective of this invention is achieved through the following technical solution: a fusion quality assessment method based on q-order orthogonal fuzzy saliency, the method comprising:

[0006] S1. Divide the medical image into blocks to obtain image sub-blocks, and then perform max pooling operation on the image sub-blocks;

[0007] S2. Perform q-order orthogonal blurring on the image sub-blocks after max pooling, and then calculate the q-order orthogonal blur proximity of the image sub-blocks.

[0008] S3. Combine the q-order orthogonal fuzzy proximity of the image sub-blocks with the evaluation index to obtain the preference score of the image sub-blocks;

[0009] S4. Obtain the saliency coefficient of the image sub-blocks according to their priority, and calculate the proximity of the image sub-blocks.

[0010] S5. Make a final selection of medical images based on the degree of proximity.

[0011] The process of dividing medical images into blocks to obtain image sub-blocks includes:

[0012] The first modality of medical source images and the second modality of medical source images are denoted as I1 and I2, respectively. The two modalities of medical source images are then fused using two different methods to obtain the first fused image F1 and the second fused image F2.

[0013] Divide I1 and I2 into p non-overlapping sub-blocks of the same size, denoted as I1=[I 11 I 12 ...,I 1i ...,I 1p ] and I2=[I 21 I 22 ,…,I 2i ...,I 2p ], where i∈{1,2,...,p}, p is the total number of sub-blocks, I 1i I represents the i-th sub-block of the medical source image I1 of the first modality. 2i This represents the i-th sub-block of the medical source image I2 in the second modality;

[0014] Divide F1 and F2 into p non-overlapping sub-blocks of the same size, denoted as F1=[B 11 B 12 ,..,B 1i ...,B 1p ] and F2=[B 21 B 22 ,..,B 2i ...,B 2p ], where B 1i B represents the i-th sub-block of the first fused image F1. 2i This represents the i-th sub-block of the second fused image F2.

[0015] The step of performing max pooling on the image sub-blocks includes:

[0016] For B 1i B 2i I 1i I 2i Extract the maximum pixel value within each pixel and perform max pooling.

[0017] The first fused image F1 maximum pooling result MB is obtained. 1i =MAX(B 1i The second type of fused image F2 max pooling result MB 2i =MAX(B 2i The maximum pooling result MI of the medical source image I1 in the first modality 1i =MAX(I 1i The second modality of medical source image I2 max pooling result MI 2i =MAX(I 2i ), where MB 1i MB represents the main features of the i-th sub-block of the first fused image F1. 2iMI represents the main features of the i-th sub-block of the second fused image F2. 1i MI represents the main features of the i-th sub-block of the medical source image I1 of the first modality. 2i The MAX(B) represents the principal feature of the i-th sub-block of the medical source image I2 in the second modality. 1i ) represents the max pooling operation on the i-th sub-block of the first fused image F1, MAX(B 2i ) represents the max pooling operation on the i-th sub-block of the second fused image F2, MAX(I 1i ) represents the max pooling operation on the i-th sub-block of the first medical source image I1, MAX(I 2i ) represents the max pooling operation of the i-th sub-block of the second type of medical source image I2.

[0018] The step of performing q-order orthogonal blurring on the image sub-blocks after max pooling includes:

[0019] For the i-th sub-block, calculate the average of its four max pooling results. Mean i This represents the arithmetic mean of the four main features of the i-th sub-block group;

[0020] Calculate the standard deviation of the four main characteristics as follows: To characterize the fluctuation amplitude of features in each image sub-block, where Std i The standard deviation of the principal features of the i-th sub-block group;

[0021] Regarding B 1i B 2i I 1i I 2i The q-th order orthogonal fuzzy membership degrees are calculated based on Gaussian functions. , , , Among them, Mem(B 1i ) q Mem(B) represents the q-th order orthogonal fuzzy membership degree of the i-th sub-block of the first fused image F1. 2i ) q Mem(I) represents the q-th order orthogonal fuzzy membership degree of the i-th sub-block of the second fused image F2. 1i ) q Mem(I) represents the q-th orthogonal fuzzy membership degree of the i-th sub-block of the medical source image I1 in the first modality. 2i ) q Let q be the orthogonal fuzzy membership degree of the i-th sub-block of the medical source image I2 in the second modality;

[0022] Based on the obtained orthogonal fuzzy membership results, calculate the q-th order orthogonal fuzzy non-membership NMem(B) of the i-th sub-block of the first fused image F1. 1i ) q =(1-Mem(B 1i ) q ω ) 1 / ω ) q The q-th order orthogonal fuzzy non-membership degree NMem(B) of the second type of fused image F2 is... 2i ) q =(1-Mem(B 2i ) q ω ) 1 / ω ) q The q-th order orthogonal fuzzy non-membership degree NMem(I) of the i-th sub-block of the medical source image I1 in the first modality 1i ) q =(1-Mem(I 1i ) q ω ) 1 / ω ) q The q-th order orthogonal fuzzy non-membership degree NMem(I) of the i-th sub-block of the medical source image I2 in the second modality 2i ) q =(1-Mem(I 2i ) q ω ) 1 / ω ) q in, For the adjustment parameters of the q-th order orthogonal fuzzy set;

[0023] Combining membership and non-membership, the q-th order orthogonal fuzzy hesitation degree of each sub-block is calculated to characterize the hesitation in feature attribution of the sub-block, which is π(B 1i ) q =1-Mem(B 1i ) q -NMem(B 1i ) q ,π(B 2i ) q =1-Mem(B 2i ) q -NMem(B 2i ) q ,π(I 1i ) q =1-Mem(I 1i ) q -NMem(I 1i )q ,π(I 2i ) q =Mem(I 2i ) q -NMem(I 2i ) q , where π(B) 1i ) q Let π(B) represent the q-th order orthogonal fuzziness hesitation of the i-th sub-block of the first fused image F1. 2i ) q π(I) represents the q-th order orthogonal ambiguity hesitation of the i-th sub-block of the second fused image F2. 1i ) q π(I) represents the q-th order orthogonal fuzziness hesitation of the i-th sub-block of the medical source image I1 in the first modality. 2i ) q The q-th order orthogonal fuzzy hesitation of the i-th sub-block of the medical source image I2 in the second modality is represented.

[0024] The calculation of the q-order orthogonal blur proximity of the image sub-blocks includes:

[0025] The proximity between the fused image sub-block and the medical source image sub-block is calculated using Euclidean distance, yielding S1(B 1i I 1i S2(B) 1i I 2i S3(B) 2i I 1i S4(B) 2i I 2i ), where S1(B 1i I 1i ) represents B 1i with I 1i The q-order orthogonal fuzzy proximity, S2(B 1i I 2i ) represents B 1i with I 2i The q-order orthogonal fuzzy proximity, S3(B 2i I 1i ) represents B 2i with I 1i The q-order orthogonal fuzzy proximity, S4(B 2i I 2i ) represents B 2i with I 2i The q-order orthogonal fuzzy proximity.

[0026] The steps in S3 include:

[0027] By combining the proximity between the fused image sub-block and the medical source image sub-block with evaluation metrics, the preference scores for the fused image sub-block are obtained as W(B). 1i )=W1S1(B 1i ,I 1i )+W2S2(B 1i ,I 2i ), W(B 2i )=W1S1(B 2i ,I 1i )+W2S2(B 2i ,I 2i ), where W1 and W2 are the weight coefficients of medical source images I1 and I2 after normalization by different evaluation indicators, respectively, W(B 1i ) represents the i-th sub-block B of the first fused image F1. 1i The overall preference score, W(B) 2i ) represents the i-th sub-block B of the second fused image F2. 2i The overall preference score is the score of the image. The higher the preference score, the better the overall fusion effect of the fused image sub-block.

[0028] The process of obtaining the saliency coefficient of image sub-blocks based on their priority sorting includes:

[0029] Based on preference scores, B 1i and B 2i According to their comprehensive preference scores W(B) 1i ) and W(B 2i Sort the data from highest to lowest, and the resulting sequence is B. 1a B 1b B 1i ... B 1f B 2a B 2b B 2i ... B 2f Among them, B 1a B represents the a-th sub-block of the first fused image F1. 1b B represents the b-th sub-block of the first fused image F1. 1i B represents the i-th sub-block of the first fused image F1. 1f B represents the f-th sub-block of the first fused image F1. 2a B represents the a-th sub-block of the second fused image F2. 2b B represents the b-th sub-block of the second fused image F2. 2irepresents the i-th sub-block of the second fused image F2, B 2f represents the f-th sub-block of the second fused image F2;

[0030] use to represent the saliency coefficient of the a-th sub-block of the first fused image F1, and set , other sub-blocks are deduced by analogy to obtain , , ..., ;

[0031] Similarly, use to represent the saliency coefficient of the a-th sub-block of the second fused image F2, set , , , ..., .

[0032] The calculating the closeness of image sub-blocks comprises:

[0033] calculating the closeness of the fused image sub-blocks based on the sorting rank, and converting the sorted fused image sub-block information into a comprehensive closeness index as , , where LA(F1) is the closeness of the first fused image F1, LA(F2) is the closeness of the second fused image F2, and a larger closeness value indicates a better overall effect of the corresponding fused image, represents the second normalization coefficient.

[0034] The step S5 comprises:

[0035] If LA(F1)≥LA(F2), select the first fused image F1;

[0036] If LA(F1)<LA(F2), select the second fused image F2.

[0037] The present invention has the following advantages: according to the fusion quality assessment method based on q-rung orthopair fuzzy saliency, the q-rung orthopair fuzzy theory and image saliency are deeply fused, so that multi-dimensional, accurate and robust assessment of the fusion quality is realized, the method is applicable to various scenes such as infrared / visible light fusion, medical multi-modal fusion, remote sensing multi-spectral fusion, etc., provides reliable support for optimization and practical application of fusion algorithms, and has a broad application prospect. Brief Description of the Drawings

[0038] Figure 1 is a schematic flowchart of the present invention. Detailed Description of the Embodiments

[0039] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application provided below with reference to the accompanying drawings is not intended to limit the scope of protection of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. The present invention will be further described below with reference to the accompanying drawings.

[0040] This invention specifically relates to a fusion quality assessment method based on q-order orthogonal fuzzy saliency. Addressing the problems of existing fusion quality assessment methods, such as single indicators, insufficient uncertainty quantification, poor robustness, and weak adaptability to visual perception, this invention first divides the medical source image and the fused image into blocks, performing max pooling on each image sub-block. Second, based on q-order orthogonal fuzzy set theory, it calculates the membership, non-membership, and hesitation degrees of each sub-block to characterize feature uncertainty. Next, it calculates the q-order orthogonal fuzzy proximity between the fused image sub-blocks and the medical source image sub-blocks, quantifying their feature similarity and reflecting the ability to retain fused information. Then, it calculates preference scores using different evaluation indicators. Next, based on the scores, it prioritizes the sub-blocks of the fusion result, obtaining the saliency coefficient of each sub-block and calculating the proximity degree. Finally, it makes the final selection based on the proximity degree.

[0041] like Figure 1 As shown, it specifically includes the following:

[0042] Step 1: Multimodal medical fusion image segmentation processing;

[0043] Let the medical source image of the first modality and the medical source image of the second modality be represented as I1 and I2, respectively. The medical source images of the two modalities are fused through two different methods to obtain the first fused image F1 and the second fused image F2.

[0044] Divide I1 and I2 into p non-overlapping sub-blocks of the same size, denoted as I1=[I 11 I 12 ...,I 1i ...,I 1p ] and I2=[I 21 I 22 ,…,I 2i ...,I 2p ], where i∈{1,2,...,p}, p is the total number of sub-blocks, I1i I represents the i-th sub-block of the medical source image I1 of the first modality. 2i This represents the i-th sub-block of the medical source image I2 in the second modality;

[0045] Divide F1 and F2 into p non-overlapping sub-blocks of the same size, denoted as F1=[B 11 B 12 ,..,B 1i ...,B 1p ] and F2=[B 21 B 22 ,..,B 2i ...,B 2p ], where B 1i B represents the i-th sub-block of the first fused image F1. 2i This represents the i-th sub-block of the second fused image F2.

[0046] Step 2: Perform max pooling on the image sub-blocks;

[0047] For B 1i B 2i I 1i I 2i The maximum pixel value within each sub-block is extracted and max pooled to characterize the main features of the sub-block.

[0048] The first fused image F1 maximum pooling result MB is obtained. 1i =MAX(B 1i );

[0049] The second type of fused image F2 max pooling result MB 2i =MAX(B 2i );

[0050] The first modality of medical source image I1 max pooling result MI 1i =MAX(I 1i );

[0051] The second modality's medical source image I2 max pooling result MI 2i =MAX(I 2i );

[0052] Among them, MB 1i MB represents the main features of the i-th sub-block of the first fused image F1. 2i MI represents the main features of the i-th sub-block of the second fused image F2. 1i MI represents the main features of the i-th sub-block of the medical source image I1 of the first modality. 2iThe MAX(B) represents the principal feature of the i-th sub-block of the medical source image I2 in the second modality. 1i ) represents the max pooling operation on the i-th sub-block of the first fused image F1, MAX(B 2i ) represents the max pooling operation on the i-th sub-block of the second fused image F2, MAX(I 1i ) represents the max pooling operation on the i-th sub-block of the first medical source image I1, MAX(I 2i ) represents the max pooling operation of the i-th sub-block of the second type of medical source image I2.

[0053] Step 3: Perform q-order orthogonal blurring on the image sub-blocks;

[0054] (1) For the i-th sub-block, calculate the average of its four max pooling results, denoted as:

[0055] ,

[0056] Among them, Mean i Represents the i-th sub-block group (B 1i B 2i ,I 1i ,I 2i The arithmetic mean of the four main features represents the average level of the main features of the current sub-block group.

[0057] (2) Calculate the standard deviation of the four main features to characterize the fluctuation range of the features of each image sub-block, denoted as:

[0058] ,

[0059] Among them, Std i For the i-th sub-block group (B 1i B 2i ,I 1i ,I 2i The standard deviation of the main characteristics of ).

[0060] (3) Regarding B 1i B 2i I 1i I 2i The q-th order orthogonal fuzzy membership degrees calculated based on the Gaussian function are as follows:

[0061] ,

[0062] ,

[0063] ,

[0064] ,

[0065] Among them, Mem(B 1i ) q Mem(B) represents the q-th order orthogonal fuzzy membership degree of the i-th sub-block of the first fused image F1. 2i ) q Mem(I) represents the q-th order orthogonal fuzzy membership degree of the i-th sub-block of the second fused image F2. 1i ) q Mem(I) represents the q-th orthogonal fuzzy membership degree of the i-th sub-block of the medical source image I1 in the first modality. 2i ) q Let q be the orthogonal fuzzy membership degree of the i-th sub-block of the medical source image I2 in the second modality.

[0066] (4) Based on the membership results, the q-order orthogonal fuzzy non-membership degrees of each sub-block are calculated as follows:

[0067] NMem(B 1i ) q =(1-Mem(B 1i ) q ω ) 1 / ω ) q，

[0068] NMem(B 2i ) q =(1-Mem(B 2i ) q ω ) 1 / ω ) q，

[0069] NMem(I 1i ) q =(1-Mem(I 1i ) q ω ) 1 / ω ) q，

[0070] NMem(I 2i ) q =(1-Mem(I 2i ) q ω ) 1 / ω ) q，

[0071] in, NMem(B) is the adjustment parameter for the q-th order orthogonal fuzzy set. 1i ) q NMem(B) represents the q-th order orthogonal fuzzy non-membership degree of the i-th sub-block of the first fused image F1. 2i) q NMem(I) represents the q-th order orthogonal fuzzy non-membership degree of the i-th sub-block of the second fused image F2. 1i ) q NMem(I) represents the q-th orthogonal fuzzy non-membership degree of the i-th sub-block of the medical source image I1 of the first modality. 2i ) q This represents the q-th orthogonal fuzzy non-membership degree of the i-th sub-block of the medical source image I2 in the second modality; and satisfies:

[0072] NMem(B 1i ) q + Mem(MB 1i ) q ≤1,

[0073] NMem(B 2i ) q + Mem(MB 2i ) q ≤1,

[0074] NMem(I 1i ) q + Mem(I 1i ) q ≤1,

[0075] NMem(I 2i ) q + Mem(I 2i ) q ≤1,

[0076] Where q≥1.

[0077] (5) Combining membership and non-membership, calculate the q-order orthogonal fuzzy hesitation degree of each sub-block to characterize the hesitation in the feature assignment of the sub-block, as follows:

[0078] π(B 1i ) q =1-Mem(B 1i ) q -NMem(B 1i ) q ,

[0079] π(B 2i ) q =1-Mem(B 2i ) q -NMem(B 2i ) q ,

[0080] π(I 1i ) q =1-Mem(I 1i) q -NMem(I 1i ) q ,

[0081] π(I 2i ) q =Mem(I 2i ) q -NMem(I 2i ) q ,

[0082] Among them, π(B) 1i ) q Let π(B) represent the q-th order orthogonal fuzziness hesitation of the i-th sub-block of the first fused image F1. 2i ) q π(I) represents the q-th order orthogonal ambiguity hesitation of the i-th sub-block of the second fused image F2. 1i ) q π(I) represents the q-th order orthogonal fuzziness hesitation of the i-th sub-block of the medical source image I1 in the first modality. 2i ) q The q-th order orthogonal fuzzy hesitation of the i-th sub-block of the medical source image I2 in the second modality is represented.

[0083] Step 4: Calculate the q-order orthogonal fuzzy proximity.

[0084] The proximity between the fused image sub-blocks and the medical source image sub-blocks is calculated using Euclidean distance, quantifying their feature similarity and reflecting the information preservation ability of the fused image. The results are as follows:

[0085] ,

[0086] ,

[0087] ,

[0088] ,

[0089] in, S1(B) represents the first normalization coefficient. 1i I 1i ) represents B 1i with I 1i The q-order orthogonal fuzzy proximity, S2(B 1i I 2i ) represents B 1i with I 2i The q-order orthogonal fuzzy proximity, S3(B 2i I 1i ) represents B 2i with I 1iThe q-order orthogonal fuzzy proximity, S4(B 2i I 2i ) represents B 2i with I 2i The q-order orthogonal fuzzy proximity.

[0090] Step 5: Calculate the preference score;

[0091] By combining the proximity between the fused image sub-blocks and the medical source image sub-blocks with evaluation metrics, the preference scores for the fused image sub-blocks are obtained as follows:

[0092] W(B 1i )=W1S1(B 1i ,I 1i )+W2S2(B 1i ,I 2i ),

[0093] W(B 2i )=W1S1(B 2i ,I 1i )+W2S2(B 2i ,I 2i ),

[0094] Where W1 and W2 represent the weight coefficients of I1 and I2 after normalization by different evaluation indicators, respectively, W(B 1i ) represents the i-th sub-block B of the first fused image F1. 1i The overall preference score, W(B) 2i ) represents the i-th sub-block B of the second fused image F2. 2i The overall preference score is the score of the image. The higher the preference score, the better the overall fusion effect of the fused image sub-block.

[0095] Step 6: Obtain the saliency coefficient of the image sub-blocks by sorting them according to their priority;

[0096] Based on preference scores, B 1i and B 2i According to their comprehensive preference scores W(B) 1i ) and W(B 2i Sort the data from highest to lowest, and the resulting sorted sequence is:

[0097] B 1a B 1b B 1i ... B 1f ,

[0098] B 2a B 2b B 2i ... B 2f ,

[0099] Among them, B 1a B represents the a-th sub-block of the first fused image F1. 1b B represents the b-th sub-block of the first fused image F1. 1i B represents the i-th sub-block of the first fused image F1. 1f B represents the f-th sub-block of the first fused image F1. 2a B represents the a-th sub-block of the second fused image F2. 2b B represents the b-th sub-block of the second fused image F2. 2i B represents the i-th sub-block of the second fused image F2. 2f This represents the f-th sub-block of the second fused image F2;

[0100] use Let represent the significance coefficient of the a-th sub-block of the first fused image F1, and let The other sub-blocks are obtained in the same way. , , ..., ;

[0101] Similarly, using Let F2 represent the significance coefficient of the a-th sub-block of the second fused image. , , , ..., .

[0102] Step 7: Calculate the proximity of image sub-blocks;

[0103] The proximity of the fused image sub-blocks is calculated based on their sorting order, and the sorted information of the fused image sub-blocks is transformed into a comprehensive proximity index:

[0104] ,

[0105] ,

[0106] Wherein, LA(F1) represents the proximity of the first fused image F1, and LA(F2) represents the proximity of the second fused image F2. Both values ​​range from [0,1]. A larger proximity value indicates a better overall effect of the corresponding fused image. This represents the second normalization coefficient.

[0107] Step 8, Final Decision;

[0108] The two fused images are finally selected based on the magnitude of the closeness LA value:

[0109] If LA(F1)≥LA(F2), the first fused image F1 is selected;

[0110] If LA(F1)<LA(F2), the second fused image F2 is selected.

[0111] The above description is merely the preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as an exclusion of other embodiments, but can be used for various other combinations, modifications and improvements, and can be modified through the above teachings or the technology or knowledge in the related art within the concept scope described herein. All alterations and modifications made by those skilled in the art that do not depart from the spirit and scope of the present invention shall fall within the protection scope of the appended claims of the present invention.

Claims

1. A fusion quality assessment method based on q-order orthogonal fuzzy saliency, characterized in that: The method comprises: S1. performing block processing on a medical image to obtain image sub-blocks, and then performing a max pooling operation on the image sub-blocks; S2. performing q-order orthogonal fuzzification on the max-pooled image sub-blocks, and then calculating the q-order orthogonal fuzzy proximity of the image sub-blocks; S3. combining the q-order orthogonal fuzzy proximity of the image sub-blocks with an evaluation index to obtain a preference score of the image sub-blocks; S4. obtaining a saliency coefficient of the image sub-blocks according to the priority ranking of the image sub-blocks, and calculating the closeness degree of the image sub-blocks; S5. performing final selection on the medical image based on the magnitude of the closeness degree.

2. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 1, characterized in that: The performing block processing on a medical image to obtain image sub-blocks comprises: representing a medical source image of a first modality and a medical source image of a second modality as I1 and I2 respectively, and obtaining a first fused image F1 and a second fused image F2 after the two medical source images of the modalities respectively pass through two fusion modes; Divide I1 and I2 into p non-overlapping sub-blocks of the same size, denoted as I1=[I 11 I 12 ...,I 1i ...,I 1p ] and I2=[I 21 I 22 ,…,I 2i ...,I 2p ], where i∈{1,2,...,p}, p is the total number of sub-blocks, I 1i I represents the i-th sub-block of the medical source image I1 of the first modality. 2i This represents the i-th sub-block of the medical source image I2 in the second modality; Divide F1 and F2 into p non-overlapping sub-blocks of the same size, denoted as F1=[B 11 B 12 ,..,B 1i ...,B 1p ] and F2=[B 21 B 22 ,..,B 2i ...,B 2p ], where B 1i B represents the i-th sub-block of the first fused image F1. 2i This represents the i-th sub-block of the second fused image F2.

3. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 2, characterized in that: The then performing a max pooling operation on the image sub-blocks comprises: For B 1i B 2i I 1i I 2i Extract the maximum pixel value within each pixel and perform max pooling. The first fused image F1 maximum pooling result MB is obtained. 1i =MAX(B 1i The second type of fused image F2 max pooling result MB 2i =MAX(B 2i The maximum pooling result MI of the medical source image I1 in the first modality 1i =MAX(I 1i The second modality of medical source image I2 max pooling result MI 2i =MAX(I 2i ), where MB 1i MB represents the main features of the i-th sub-block of the first fused image F1. 2i MI represents the main features of the i-th sub-block of the second fused image F2. 1i MI represents the main features of the i-th sub-block of the medical source image I1 of the first modality. 2i The MAX(B) represents the principal feature of the i-th sub-block of the medical source image I2 in the second modality. 1i ) represents the max pooling operation on the i-th sub-block of the first fused image F1, MAX(B 2i ) represents the max pooling operation on the i-th sub-block of the second fused image F2, MAX(I 1i ) represents the max pooling operation on the i-th sub-block of the first medical source image I1, MAX(I 2i ) represents the max pooling operation of the i-th sub-block of the second type of medical source image I2.

4. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 3, characterized in that: The performing q-order orthogonal fuzzification on the max-pooled image sub-blocks comprises: For the i-th sub-block, calculate the average of its four max pooling results. Mean i This represents the arithmetic mean of the four main features of the i-th sub-block group; Calculate the standard deviation of the four main characteristics as follows: To characterize the fluctuation amplitude of features in each image sub-block, where Std i The standard deviation of the principal features of the i-th sub-block group; Regarding B 1i B 2i I 1i I 2i The q-th order orthogonal fuzzy membership degrees are calculated based on Gaussian functions. , , , Among them, Mem(B 1i ) q Mem(B) represents the q-th order orthogonal fuzzy membership degree of the i-th sub-block of the first fused image F1. 2i ) q Mem(I) represents the q-th order orthogonal fuzzy membership degree of the i-th sub-block of the second fused image F2. 1i ) q Mem(I) represents the q-th orthogonal fuzzy membership degree of the i-th sub-block of the medical source image I1 in the first modality. 2i ) q Let q be the orthogonal fuzzy membership degree of the i-th sub-block of the medical source image I2 in the second modality; Based on the obtained orthogonal fuzzy membership results, calculate the q-th order orthogonal fuzzy non-membership NMem(B) of the i-th sub-block of the first fused image F1. 1i ) q =(1-Mem(B 1i ) q ω ) 1 / ω ) q The q-th order orthogonal fuzzy non-membership degree NMem(B) of the second type of fused image F2 is... 2i ) q =(1-Mem(B 2i ) q ω ) 1 / ω ) q The q-th order orthogonal fuzzy non-membership degree NMem(I) of the i-th sub-block of the medical source image I1 in the first modality 1i ) q =(1-Mem(I 1i ) q ω ) 1 / ω ) q The q-th order orthogonal fuzzy non-membership degree NMem(I) of the i-th sub-block of the medical source image I2 in the second modality 2i ) q =(1-Mem(I 2i ) q ω ) 1 / ω ) q ,in, For the adjustment parameters of the q-th order orthogonal fuzzy set; Combining membership and non-membership, the q-th order orthogonal fuzzy hesitation degree of each sub-block is calculated to characterize the hesitation in feature attribution of the sub-block, which is π(B 1i ) q =1-Mem(B 1i ) q -NMem(B 1i ) q ,π(B 2i ) q =1-Mem(B 2i ) q -NMem(B 2i ) q ,π(I 1i ) q =1-Mem(I 1i ) q -NMem(I 1i ) q ,π(I 2i ) q =Mem(I 2i ) q -NMem(I 2i ) q , where π(B) 1i ) q Let π(B) represent the q-th order orthogonal fuzziness hesitation of the i-th sub-block of the first fused image F1. 2i ) q π(I) represents the q-th order orthogonal ambiguity hesitation of the i-th sub-block of the second fused image F2. 1i ) q π(I) represents the q-th order orthogonal fuzziness hesitation of the i-th sub-block of the medical source image I1 in the first modality. 2i ) q The q-th order orthogonal fuzzy hesitation of the i-th sub-block of the medical source image I2 in the second modality is represented.

5. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 4, characterized in that: The calculating the q-order orthogonal fuzzy proximity of the image sub-blocks comprises: The proximity between the fused image sub-block and the medical source image sub-block is calculated using Euclidean distance, yielding S1(B 1i I 1i S2(B) 1i I 2i S3(B) 2i I 1i S4(B) 2i I 2i ), where S1(B 1i I 1i ) represents B 1i with I 1i The q-order orthogonal fuzzy proximity, S2(B 1i I 2i ) represents B 1i with I 2i The q-order orthogonal fuzzy proximity, S3(B 2i I 1i ) represents B 2i with I 1i The q-order orthogonal fuzzy proximity, S4(B 2i I 2i ) represents B 2i with I 2i The q-order orthogonal fuzzy proximity.

6. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 5, characterized in that: The step S3 comprises: By combining the proximity between the fused image sub-block and the medical source image sub-block with evaluation metrics, the preference scores for the fused image sub-block are obtained as W(B). 1i )=W1S1(B 1i ,I 1i )+W2S2(B 1i ,I 2i ), W(B 2i )=W1S1(B 2i ,I 1i )+W2S2(B 2i ,I 2i ), where W1 and W2 are the weight coefficients of medical source images I1 and I2 after normalization by different evaluation indicators, respectively, W(B 1i ) represents the i-th sub-block B of the first fused image F1. 1i The overall preference score, W(B) 2i ) represents the i-th sub-block B of the second fused image F2. 2i The overall preference score is the score of the image. The higher the preference score, the better the overall fusion effect of the fused image sub-block.

7. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 6, characterized in that: The obtaining a saliency coefficient of the image sub-blocks according to the priority ranking of the image sub-blocks comprises: Based on preference scores, B 1i and B 2i According to their comprehensive preference scores W(B) 1i ) and W(B 2i Sort the data from highest to lowest, and the resulting sequence is B. 1a B 1b B 1i ... B 1f B 2a B 2b B 2i ... B 2f Among them, B 1a B represents the a-th sub-block of the first fused image F1. 1b B represents the b-th sub-block of the first fused image F1. 1i B represents the i-th sub-block of the first fused image F1. 1f B represents the f-th sub-block of the first fused image F1. 2a B represents the a-th sub-block of the second fused image F2. 2b B represents the b-th sub-block of the second fused image F2. 2i B represents the i-th sub-block of the second fused image F2. 2f This represents the f-th sub-block of the second fused image F2; use Let represent the significance coefficient of the a-th sub-block of the first fused image F1, and let The other sub-blocks are obtained in the same way. , , ..., ; Similarly, using Let F2 represent the significance coefficient of the a-th sub-block of the second fused image. , , , ..., .

8. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 7, characterized in that: The calculating the closeness degree of the image sub-blocks comprises: The proximity of fused image sub-blocks is calculated based on their ranking, and the information of the ranked fused image sub-blocks is transformed into a comprehensive proximity index. , Where LA(F1) represents the proximity of the first fused image F1, and LA(F2) represents the proximity of the second fused image F2. A larger proximity value indicates a better overall effect of the corresponding fused image. This represents the second normalization coefficient.

9. The fusion quality assessment method based on q-order orthogonal fuzzy saliency according to claim 8, characterized in that: The step S5 comprises: if LA(F1)≥LA(F2), selecting the first fused image F1; if LA(F1)<LA(F2), selecting the second fused image F2.

Images