Research On Image Fusion And Super Resolution Reconstruction Algorithm In The Nonsubsampled Dual-tree Complex Contourlet Transform Domain

Image fusion is eager to get a composite image which integrates the complementary and redundant information of multiple images data of the same scene, so that the composite image contains better and more accurate description of the scene; The objective of image super-resolution reconstruction is to obtain a relatively high resolution image from one or multiple low resolution images with complementary information. Image fusion and super-resolution reconstruction are two preprocessing of image processing, which have an important influence on the subsequent image detection, texture analysis, feature extraction and pattern recognition, etc. Therefore, the research of image fusion and super-resolution reconstruction has important significance.Wavelet transform is good at time-frequency localization, which is widely applied in the field of image processing. However, the traditional wavelet transform is shift sensitive and lacks direction selectivity, to overcome the disadvantages of the traditional wavelet transform, some new types of image sparse representation tools that can more sparsely represent the local characteristics of image are proposed in recent years. In this thesis, some new types of image sparse representation tools such as nonsubsampled dual-tree complex contourlet transform and nonsubsampled quaternion contourlet transform are preliminarily studied, and combined with advantages of compressive sensing, sparse representation and pulse coupled neural network, then the image fusion and super-resolution reconstruction applications based on these new types of image sparse representation tools are in-depth studied. The main work is as follows:1. A novel image fusion algorithm based on nonsubsampled dual-tree complex contourlet transform(NSDTCT) and compressive sensing pulse coupled neural network(CS-PCNN) is proposed. Firstly, decompose the source images by NSDTCT to obtain the low frequency sub-band coefficients and high frequency sub-band coefficients. For the low frequency sub-band coefficients, an adaptive weighted fusion method combining the regional average gradient, regional energy with Sigmoid function is presented. For the high frequency sub-band coefficients with large amount of data, a fusion rule based on the theory of CS-PCNN is presented, and the novel sum-modified Laplacian is used for the external input of PCNN. Finally, the fused image is obtained by performing the inverse NSDTCT on the fused coefficients. The experimental results show that the proposed algorithm can improve the computation efficiency and the quality of the fused image, and outperforms other classical fusion algorithms in terms of both visual quality and objective evaluation.2. A novel remote sensing image fusion algorithm based on nonsubsampled dual-tree complex contourlet transform and sparse representation is proposed. According to the different characteristics of the low and high frequency coefficients, for the low frequency coefficients, a fusion method based on sparse representation is presented, and the fused coefficient is obtained by combining with spatial frequency and l1-norm maximum. For the high frequency coefficients, the sum-modified Laplacian is used for the external input of pulse coupled neural network(PCNN), and a fusion method based on the theory of improved PCNN is presented. The experimental results show that the proposed algorithm can improve the spatial resolution and better maintain the spectral characteristics, and outperforms other classical remote sensing fusion algorithms in terms of both visual quality and objective evaluation.3. A novel single-image super-resolution reconstruction algorithm based on interpolation and nonsubsampled quaternion contourlet transform(NSQCT) fusion is proposed. Firstly, the source image is separately interpolated by soft-decision adaptive interpolation and cubic spline interpolation. Then the NSQCT is adopted to decompose the two interpolated images, for the low frequency coefficients, an adaptive weighted fusion method combining the regional average gradient, regional energy with Sigmoid function is presented. For the high frequency coefficients, a fusion rule based on the new Sum-modified Laplacian combined with the weighted analysis is presented. Finally, the high resolution image is obtained by performing the inverse NSQCT on the fused coefficients. The experimental results show that the proposed algorithm outperforms those classical reconstruction algorithms in peak signal-to-noise ratio, structural similarity as well as visual quality.