Sparse representation uses the linear combination of overcomplete dictionary and sparse coefficients to realize the reconstruction, compression, description and representation of general signals or images, which is an efficient and accurate method. In sparse representation, the process of computing and obtaining an over-complete dictionary D is called dictionary construction, and the process of computing and obtaining sparse coefficients is called sparse coding. How accurate the sparse representation is is closely related to both.

 

Sparse representation works well for fusion of multi-modal and multi-focus images. General procedure for fusion of thermal infrared and visible images based on sparse representation. First, use the sliding window strategy to decompose each source image into multiple overlapping image blocks; next, build a dictionary by learning over-complete features of a large number of image samples; then, perform sparse coding on each image block to obtain the image block The coefficients are sparsely represented, and the coefficients are fused according to appropriate rules; finally, the over-complete words that have been learned are used. The code reconstructs the fused sparse coefficients to obtain the fused image.