Study On Algorithm Of Image Smoothing Based On Partial Differential Equation

In the course of filming and transmission, images are often mixed with a lot of noise. Noise image not only affects seriously visual effects of image, but also brings certain difficulties to the subsequent image analysis and understanding in the image processing. So image smoothing is one of the main study in image reprocess, and the smoothing quality affects directly the following process. In the smoothing process, it is contradictive to keep image character and eliminate noise. The contradiction can been solved better by image smoothing technique based on partial differential equation, especially anisotropic diffusion equation, which can smooth image selectively. The image smoothing algorithm based on partial differential equation is studied in this thesis, and the detailed study is summaried in the following:(1) Several classical image smoothing models based on partial differential equation are introduced, the experiment results of these image smoothing models are showed correspondly. A novel diffusion coefficient is introduced with an analysis of diffusion mechanisms of P-M model, then the P-M model is improved. The improved P-M model not only improves smoothing quality but also speeds up the smoothing velocity.(2) P-M model can remove noise in intra-region sufficiently but can't eliminate noise at edges successfully and line-like structure can't be held very well. Coherence-enhancing diffusion(CED)model can remove noise at edges and preserve edges very well but false texture arise in intra-region. So an image smoothing model is proposed combination of P-M model and CED model in wavelet region. After one level wavalet decomposing of image, the low frequency sub-image is smoothed based on P-M model and the other frequency sub-image are smoothed based on CED model. The experiment results show that this method is effective to image smoothing.(3) A novel adaptive total variation smoothing model is proposed with an analysis of nonstandard TV model. The method can overcome the "staircasing" effect and keep the texture of image effectively. The experiment results show that this method is effective.