Research On Remote Sensing Image Fusion Based On Non-aliasing Contourlet Transform

The image fusion technology is an important technology in the field of image processing. It makes the different kinds of sensors which obtain a variety of information characteristics in the same scene, through a certain algorithm, produce more comprehensive, accurate and stable new image than the original image information. The new image contains complementary and redundant information of the original image. In recent years, the image fusion technology has been widely used in remote sensing image, medical image, automatic target recognition, machine vision and many other fields. Remote sensing image fusion technology research has become the priorities and hot spots in the field of image fusion technology. Among unsolved key problems of remote sensing image fusion, we mainly focus on the three key technologies of them: Image interpolation technology for image fusion preprocessing, multi-spectral image and panchromatic image fusion technology and Infrared image and visible light image fusion technology.The main work of this paper is summarized as follows:(1) It introduces the basic theory and algorithm of Contourlet transform (CT) and non-aliasing Contourlet transform (NACT) in details and analyses the frequency aliasing problems of the Contourlet transform, meanwhile, it compares two kinds of transformations.(2) On the half of image fusion preprocessing, bi-linear interpolation algorithm based on non-aliasing Contourlet transform is presented, which is applied in remote sensing image fusion preprocessing. The algorithm uses bi-linear interpolation algorithm to anticipate the high frequency part on high resolution from the high frequency part on low resolution, and obtains interpolation image with higher resolution ratio than the original image by inverse NACT. The experimental results show that, compared with traditional interpolation methods, the algorithm can get higher PSNR and better image detail effect.(3) In view of multi-spectral image and panchromatic image's fusion problem, a novel image fusion algorithm based lαβspace and non-aliasing Contourlet transform is proposed in this paper for multi-spectral and panchromatic images. Firstly, the multi-spectral image is transformed into lαβspace, and the original l component and panchromatic image are decomposed into a low frequency subband image and high frequency subband images are fused with different fusion rules, respectively, and the fused coefficients are obtained. The Cycle spinning is introduced to decrease the shift-variance of the transform. Finally, the fused multi-spectral image with high spatial resolution is generated by the inverse NACT and inverse lαβtransform. Experimental results show that the accuracy and validity of this paper's method, the image fusion effect and integration of statistical indicators are better than the method of traditional IHS method; wavelet fusion method and Contourlet transform based method.(4) Arming at the infrared and visible image fusion problem, a fusion algorithm based on the improved Pulse Coupled Neural Network (PCNN) and NACT is proposed. PCNN has the advantages of global coupling and pulse synchronization, which makes PCNN can extract information from the image's edge, contour and texture. However, the PCNN model parameters are more sensitive to parameter, whose adjustment is more troublesome, so it affects the application in image fusion. The algorithm uses an improved PCNN to solve these problems with non-aliasing Contourlet transform. The algorithm is applied to the fusion of infrared image and visible light image. The experimental results show that, the method of fusion effect is better than that of Laplacian Pyramid method, wavelet transform and Contourlet transform method.(5) This paper studies the NMF theory application in image fusion, and proposes two image fusion algorithms which are based on NMF. Firstly, in view of infrared image and visible image fusion, the paper proposes an infrared image and visible image fusion algorithm which is based on the NMF and local energy. The algorithm takes full account of the NMF algorithm should extract the overall characteristics of image, meanwhile, applies it to image fusion, so the fusion effect is superior to the traditional fusion algorithm. Secondly, in view of the image fusion for multi spectral image and panchromatic, the paper proposes a multi-spectral image and panchromatic image fusion algorithm which is based on NMF and NACT. The algorithm uses the NMF algorithm for multi-spectral image pre-processing to obtain high image quality. The experimental simulation results prove the correctness and effectiveness of the algorithm. Both the subjective visual effect and the detachment condition's parameter appraisal analysis of the fusion image outperform other traditional fusion algorithm.