Research On Image Denoising Based On Wavelet And Its Statistical Model

There is always much noise in image,in order to further image analysis and communication,we need to reduce noise and improve image quality,in the image pre-processing must be de-noising. Scholars have been studying to find an effectively method to reduce the noise, but also good to retain information with image edge and texture.In recent years, with the continuous improvement of wavelet theory, with its good time-frequency localization features, the characteristics of scale changes and directivity, wavelet analysis has been applied to various disciplines. Similarly, it has also made a wide range of applications in the wavelet de-noising, and proposed a number of wavelet image de-noising algorithms.The research content of this dissertation includes the following three aspects: first aspect is introducing and studying the wavelet shrinkage threshold method and proposing a new threshold shrinkage function; second aspect is the researching on the relevant characteristics of the statistical model of wavelet coefficients, focusing on the bivariate model, combined with the inner layer and the inter layer proposed an adaptive correlation select neighborhood window with bivariate model wavelet de-noising algorithm; third aspect is the preliminary researching on the DTCWT and applying adaptive select neighborhood window with the bivariate model on image de-noising.Donoho proposed a threshold on wavelet de-noising, this paper introduced the threshold and other hard and soft threshold function, we propose a new shrinkage parameter adaptive threshold function, it can be obtained through the analysis of experimental results better effect.there is a strong correlation between wavelet coefficients,Including the inner layer and the inter layer correlation.Consider the correlation between wavelet coefficients, using a statistical model of wavelet coefficients for image de-noising is another new method of wavelet de-noising.Taking into account the inner-layer correlation of wavelet coefficients,this paper described the bivariate model in inter-layer correlation of wavelet coefficients combined with inter layer correlation, proposed a local adaptive algorithm to select the neighborhood window with bivariate model. The algorithm is compared with the traditional bivariate model is good to reduce image noise and improve the peak signal to noise ratio of the de-noised image while better retain the image edge details and other important information.Wavelet shrinkage de-noising and statistical models de-noising can been able to achieve good results is that the wavelet transform provides sparse image representation. However, the traditional discrete wavelet transform to extract does not have a good translation invariance and good directional selectivity material defects. Based on this, this paper introduces the dual tree complex wavelet transform, which not only has a local time-frequency characteristics and the multi-resolution features,but also has features such as orientation and translation invariance. On this basis, the dual tree complex wavelet transform is applied to image de-noising with bivariate model, combined with local select neighborhood window adaptive algorithm. The experimental results showed that, compared with conventional extraction of discrete wavelet transform, image de-noising algorithm based on dual tree complex wavelet transform can achieve the better de-noising effect.