Image Fusion Based On Compressed Sensing

Compressed Sensing(CS) technology is a new compression sampling technology, which can reduce the sampling data and reduce the storage and transmission. So it has broad application prospects in many fields, especially in image processing. In this paper, the compressed sensing is used for image fusion, the main work is as follows:(1) A compressed sensing image fusion method based on joint sparse representation is proposed. Because the two images are the record of the same scene, it has public information, but because of the different imaging principle, it will show different images information, which is the theoretical basis of the joint sparse model(JSM-1), so JSM-1 can be used to do image sparse representation. Firstly, the redundant dictionary is obtained with K-SVD training, and then the public sparse coefficients and independent sparse coefficients are extracted from the image, and then the coefficients are reduced by using the weighted norm. In addition, because of the strong noise suppression ability of K-SVD, the algorithm can achieve high quality image fusion and image denoising.(2)A compressed sensing image fusion method based on auto-correlation function measure(AFM) is proposed. This method is mainly for the multi-focus image fusion, from the imaging principle analysis, the focus of good image is rich in high frequency information, and from the perspective of frequency to explain the image clear and fuzzy is intuitive, is consistent with the human visual characteristics and subjective feelings. According to the Wiener Khinchin theorem, the image of the auto-correlation function and power spectrum is the Fourier transform of the auto-correlation function, so it can be through the image to evaluate the clarity of the image. First, the sparse coefficients of the two images of the same image with different focus are obtained by FFT. Then, Then, two observation vectors are obtained by the double-star sampling mode, fusion criterion based on the auto-correlation function measure(AFM) is used to fuse these two observation vectors.Finally, the sparse coefficients are reconstructed by using the adaptive gradient projection algorithm, and then the fusion image is reconstructed by IFFT. The experimental results show that the AFM fusion criterion has higher fusion image quality.(3) A adaptive gradient projection(ASPM) reconstruction algorithm is proposed. In this paper, The adaptive projection subgradient algorithm was improved by adaptive expansion coefficient adjustment mechanism.The algorithm can improve the convergence speed and precision of the algorithm. Experiments show that, compared with gradient projection and OMP algorithm,ASPM has more obvious advantages in operating time, MSE, anti noise interference.