Image fusion method based on adaptive group structure sparse dictionary learning

Abstract

The invention discloses an image fusion method based on adaptive group structure sparse dictionary learning. The method comprises concrete steps of inputting to-be-fused images and carrying out sliding window partitioning, extracting feature vectors to obtain a training sample, training an adaptive group structure dictionary, calculating the sparse representation vector of each image block, taking the corresponding sparse representation vectors of the same position and using the group structure-based L1-norm optimization method to obtain a fused coefficient expression vector matrix, finally adding the corresponding mean value to the fused coefficient expression vector matrix and performing image representation to obtain the image block, and using the reverse operation of the sliding window to obtain the final fusion image. The fused image is more accurate, the details are more abundant, and the spatial incontinuity of the fused image can be reduced.

Description

Technical field [0001] The invention belongs to the field of image processing technology, and particularly relates to image fusion, and can be used in image processing, computer vision, medical image processing, military and other fields. Background technique [0002] In recent years, sparse representation has received more and more attention as an effective method. It has been applied to many problems in image processing, such as image denoising, image fusion, and image compression. A scene contains a wealth of information, and a single imaging sensor system cannot capture all the information in the scene. In order to solve this problem, a multi-sensor image fusion system came into being. The multi-sensor image fusion system effectively utilizes the complementarity between different imaging sensors. It eliminates the redundant information between multi-sensor images and integrates the images acquired by different imaging sensors to form a more complete, clear and accurate descri...

AI Technical Summary

Problems solved by technology

However, the K-SVD algorithm has the disadvantages of constructing a dictionary: the representation ability of a single dictionary is limited, and the ability to represent the details of the image is insufficient, which will result in smoother images after fusion and poor reconstruction ability for details.

[0006] (1) Some significant information needs to be preserved in the image block with a smaller L1 norm, and such a fusion rule will not obtain this part of information in the fused image, making the information loss of the fused image relatively large. Big, not rich enough detail

[0007] (2) The rule with the largest L1 norm will cause spatial discontinuity in the fusion image

Images