Research On The Application Of Image Fusion In Image De-noising

In the process of acquisition, quantization, encoding and transmission of image, image signal will be affected by physical conditions and environment factors, and can be ineluctability disturbed by noise. The noise in image signal is an important factor that causes the image quality reduction and greatly influences high level processing such as image analysis, feature extraction and pattern recognition. So image de-noising is one of the most important content in image processing field.The principle of wavelet transform and the wavelet transform based de-noising algorithm are introduced in detail at first. According to the defect that wavelet transform cannot optimally represent curve singularity, two new multi-scale geometric analysis methods——Ridgelet and Curvelet transform are introduced and analyzed. Then we introduce basic principles and realization methods of the undecimated discrete wavelet transform (UDWT) as well as fast discrete curvelet transform (FDCT). The de-noising algorithms based on UDWT and FDCT are also given in this thesis. By comparing the simulation results of several wavelet based de-noising methods and FDCT based de-noising methods as well as combining with image fusion technology, two new image combined de-noising methods based on pixel-level image fusion are proposed.The main purpose of pixel-level image fusion based de-noising method is to remove the artifact generated by Curvelet transform so as to extend the application range of FDCT based de-noising algorithms. Wavelet fusion is first applied to the process of image de-noising in this thesis and the wavelet fusion based image combined de-noising method is also proposed. This method chooses undecimated discrete wavelet transform as an auxiliary to process homogeneous region in the image and transfers the two pictures de-noised by UDWT and FDCT algorithms into frequency domain using wavelet transform. Different fusion rules based on local energy features are adopted in low and high frequency portions to attain the purpose of weakening artifacts. The experiment results demonstrate that this combined de-nosing method can effectively remove noise in image and inhibit artifacts to some extent. The de-noised image not only retains detail information but also has good gliding property and visual effect.Considering that there are some differences existing in the de-noised results of UDWT and FDCT algorithms, quadtree decomposition is applied to the process of image de-noising and another new image combined de-noising method based on quadtree decomposition is proposed in this thesis with the purpose of inhibiting artifacts more effectively. As a powerful image structural analysis tool, quadtree decomposition can be adopted to split the image de-noised by FDCT into homogeneous regions and edge regions. Applying different fusion coefficients to those regions, artifacts can be undermined after fusion. The experiment results demonstrate that this combined de-nosing method can eliminate artifacts more effectively and has higher PSNR compared with other image de-noising algorithms. We can also apply this method to process images needing high accuracy such as medical and SAR satellite images and get better effects. The experiments indicate that the pixel-level image fusion based combined de-noising method has a wider application prospect.