Research Of Multi-Sensor Image Fusion Based On Block Sparse Representation

The continuous development of image sensor technology has brought massive, rich and varied images. There is a large degree of complementarity and redundancy among the image information of the same scenario captured by different sensors. To understand, store, process the multiple images directly will increase the difficulty and complexity of the system processing, and reduce storage efficiency. In order to solve this problem, image fusion technology came into being. The so-called image fusion technology is a process of fusing the multi-sensor images into a single image through some certain fusion methods. The fused image can not only reduce the redundancy, save the storage space to a large extent; but also enhance the complementarity and reliability of the image, make it easier to comprehensively understand the scenario and computer processing.Over the past nearly 20 years, signal sparse representation has been a very interesting research area in the signal processing filed. The purpose of signal sparse representation is to use as few atoms as possible to represent the signal with a given over-complete dictionary, getting a more succinct and effective representation of the signal. The hotspots of signal sparse representation mainly concentrated in three aspects: the sparse decomposition algorithm, the learning method of over-complete dictionary, the application of the sparse representation.The research content of this paper is multi-sensor image fusion based on the sparse representation, mainly carries on three aspects of the work as follows:(1)The two main tasks of signal sparse representation are dictionary learning and the signal’s sparse decomposition under the dictionary. Orthogonal Matching Pursuit (OMP) is a classical sparse decomposition algorithm, but this algorithm has the disadvantage of repeated computation and only select one atom at each iteration. Aiming at this problem, inverse optimization OMP algorithm is proposed. The experimental results show that the efficiency of the inverse optimization OMP algorithm is improved about 50%compared with SOMP(Simultaneous-OMP).(2)When image fusion is carried out, the amount of useful information extracted from the source image is one of the important indexes to measure whether a fusion method or model is good or not. The more useful information is, the better the quality of the fusion image is, and vice versa. In order to fuse the information of the source image as much as possible, and avoid distortion, this paper proposes a joint sparse subspace recovery (JSSR) fusion model. Experiments show that the model has achieved good result in both visual and objective evaluation.(3)Image fusion rule is another important factor affecting the quality of fusion image. The fused image with "choose max" rule tends to be over sharp and less smooth; using the "weighted average" blurs the details and reduces contrast of the image. To solve these two problems, this paper proposes unbiased fusion rule; the experiment proves that the rule has achieved good results.