Research On Multi-focus Image Fusion Based On Multi-resolution Analysis

Image fusion is to synthesize several images with the exploitation of complementary date derived from multisensor to obtain a new image, which provides richer visual information than the original images. Multi-focus image fusion is an important branch of image fusion technology. It can fuse several images which are got by each target focused separately and shot repeatedly. The purposes is getting a clear image in which each target is distinct. It is widely used in military and civilian areas such as machine vision, digital camera and object recognition and so on. Image fusion method can be categorized as three levels, pixel level fusion, feature level fusion and the decision-making level fusion. The pixel-level image fusion approaches are researched frequently at present. In pixel level image fusion, the multi-resolution image fusion algorithm is very important and widely utilized.In this paper, on the basis of pyramid transform and wavelet transform, we have studied and researched pixel-level multi-focus image fusion techniques and pretreatment before fusion, included image denoising and image registration. The dissertation consists of five chapters.In chapter one, we introduce the basic concept of multi-focus image fusion, levels of image fusion firstly. Then we present the evolution and the current research situation of image fusion based on multi-resolution.In chapter two, the elementary theory of multi-resolution included of pyramid transform and wavelet analysis theories are introduced. We focus on Mallat decomposition and reconstruction algorithm and its filters implementation process.In chapter three, the structure of pixel-level image fusion based on multi-resolution is introduced firstly. And then we present some traditional fusion methods applied to both pyramid transform and wavelet transform. By analysis and improving the existing multi-focus image fusion algorithms, we propose a new fusion approach based on pixel-clarity. We define a concept of pixel-clarity on the basis of SWT coefficients and select fused coefficients according to all pixel-clarity. Finally, we implement consistency test by morphological operation. We experiment much times to get two optimal windows and prove the excellent performance of the proposed approach. In chapter four, we discuss the pretreatment of image fusion, included image denosing and image registration. Firstly, considering both the direction selectivity of sub-band filter of wavelet decomposition and the edge and texture direction of the original image, combined of Wiener filtering, we propose a image denoising algorithm using local wiener filtering with block-adaptive windows in wavelet domain. Secondly, we discuss the image registration method based on maximum mutual information measure in detail. We introduce the affine transform and its matrix expression, bilinear interpolation of grayscale, mutual information measure, joint histogram and its Partial Volume calculation approach and using discrete wavelet transform to determine the initial value of the optimization algorithm and improve the image registration accuracy. Finally, we accomplish the approach and get good results.In chapter five, a brief summary of the dissertation is given. The suggestion for future research related to the image fusion is brought forward.