The Method Research Of Image Fusion Based On Compressed Sensing

Image fusion is an important branch and a research hotspot of the information fusion.With the rapid development of information technology, the people to the requirements of the quality of the image is getting higher and higher. This caused great pressure for signal sampling, transfer and storage. How to alleviate the pressure and can effectively extract the useful information bearing in signal have became one of the problem urgent need to resolve in signal and information processing field.The traditional image fusion method due to the great amount of data has been difficult to meet the need of the actual, so desperately need to seek new theory method.Brought in2006, the new theory,Compressive Sensing(CS) provides a new idea to solve the problem.CS is a new kind of sampling theory frame based on information.It broke through the bottleneck of the traditional Nyquist Sampling Theorem,and availed direct high-resolution signal acquisition,yet its application study in image fusion has just been initialized.Under the framework of multi-scale transformation image fusion, combined with CS theory,we complete the work as follow:1. A CS based image fusion algorithm with average gradient weighted for multi-focus image is proposed. Through to the source image region and calculation of average gradient in each image sub-block to determine its fusion weights value, can adaptive to choose the best image sub-block fusion weight value. The experimental results show that the method can obtain promising fusion performance with a low computational complexity for multi-focus image.2. A CS based image fusion algorithm with single layer wavelet transform is proposed.we take single layer wavelet decomposition for source image and only measured the high-pass wavelet coefficients of the image. Then,fused the low-pass wavelet coefficients and the measurements of high-pass wavelet coefficients with different schemes. Finally, the composite coefficients of the fused image are subsequently recovered via the reconstruction algorithm.The experimental results validate that the performance of the proposed approach in both evaluation indexes and visual quality.3. A CS based image fusion algorithm with Nonsubsampled Contourlet Transform(NSCT) is proposed. Specifically, in order to design a better fusion rule, we introduce the structural similarity (SSIM) to extract the data relation between the measurements of two high-pass sub-images decomposition by NSCT of original images. The experimental results show that the proposed method can get similar effect compare with the traditional NSCT based image fusion method with few sampling data.This research is supported by the National Natural Science Foundation of China (No:61072106),the Fundamental Research Funds for the Central Universities (No.JY10000902045) and the Fund for Foreign Scholars in University Research and Teaching Programs (the111Project)(No:B07048).