Based On The Wavelet Image Denoising And Finite Ridgelet Transform

Image de-noising is a process which preserves important image information and removes image noise as much as possible. Image de-noising is a necessary step of image analysis, whose performance is directly dependent on the performance of image de-noising, so image de-noising is one of the most important aspects in the image processing field. The wavelet de-noising method is a kind of image de-noising methods, which can achieve good de-noising performance. But the wavelet transform is not good at processing line singularities or surface singularities, so researchers have proposed the finite ridgelet transform to improve the default of the wavelet transform.The paper will improve the de-noising performance of the wavelet de-noising methods from two aspects. One is the relation of wavelet coefficients, and the other is that the wavelet can’t process line singularities well.The paper focuses on the faults of the method considering the neighborhood dependencies of the wavelet coefficients and the method considering the inter-scale dependencies of the wavelet coefficients, and then proposes a de-noising method, which takes the inter-scale and neighborhood dependencies of wavelet coefficients into account. The proposed method uses the correlation of wavelet coefficients and the average magnitudes of the surrounding wavelet coefficients within a local window to describe the intra-scale and neighborhood dependencies of wavelet coefficients respectively. Based on the two dependencies, the paper proposes a new threshold function and researches the relation between the threshold and image noise standard deviation. Experiments show that compared with the method considering the neighborhood dependencies of the wavelet coefficients and the method considering the inter-scale dependencies of the wavelet coefficients, the proposed method can achieve higher PSNRs.In order to improve the fault that the wavelet transform can not process line singularities well, the paper proposes an image de-noising method combining the wavelet and the finite ridgelet transforms. First, the proposed method divides the whole image into image blocks, and then the homogenous blocks and non-homogenous blocks are selected and processed by the wavelet de-noising method and the finite ridgelet de-noising method, respectively. Based on the histogram of the standard deviation of image blocks, the paper proposes a selection rule of the homogenous blocks. Compared with the wavelet de-noising method using Bayes rule and the finite ridgelet de-noising method with Bayes rule, the de-noising performance of the proposed method is better. When the added noise standard deviation is above20, the SNRs of the proposed method can increase by1db.