A Study On Image Wavelet Denosing Method Basing On Region Of Interest

A demand for static image denoising was proposed as Multimedia andInternet quickly developed. The traditional image denoising methods were basedon the entire image, while the standard JPEG2000 proposed a coding functionbased on ROI, which allows a given part of the image (the part that you areinterested in)coded at a better quantity. It also shows JPEG2000's characteristicthat coding are based on content. This idea is based on the phenomenon thatpeople's interest in an image varies according to their psychologies, culturebackgrounds, circumstance and the various applications.Moreover, the traditional denoising methods don't take into considerationthe differences among various image categories of image, even denoised with asingle method. It will result in a contradictory between preserving intact detailsand removing maximum noise.In this paper, A new wavelet denoising method based on ROI is proposed. Inorder to improve PSNR (Peak Signal-Noise Ratio) in the whole image, thismethod processes the chosen ROI and background differently while meeting thevision features of human eyes.Researching content :First, in this paper, we investigate image classification. In this paper, weadopt a simple but efficient classification algorithm, which devides the image intosmooth regions and texture regions with the following principle: In smoothregions, the intensity of pixels changes smoothly, therefore, the partial varianceall over the region should be added up to zero, or be the variance of noise if evercontained. While in texture regions, the intensity of pixels changes fiercely,resulting in shape differences between partial variance and the variance of noisecontained. Experiments on Standard images such as Baboon, San Diego, Lena,woman, peppers have shown that the proposed method has a remarkably superiorability to achieve better visual quality while preserving the edge information. Second, in this paper, also investigate the denoising efficiencies of variousvisushrink upon different kinds of images. The input image is affected by additive,stationary noise. The noise can be colored or white. It can be proved that thewavelet transform of stationary noise is stationary at each resolution level andwithin erery component (vertical, horizontal, and diagonal). The noise should notbe "too large". In that case, the noise has a relatively small influence on theimportant large clean coefficients. But the signal has a great influence. So we canchoose an optimal threshold for subband, the small coefficients are replacedbyzero because they are dominated by noise and carry only a small amount ofinformation. The coefficiencies above the threshold are remained. At last, thecoefficients should be reconstruction. Taking San Diego and Lena as examples,we use a range of wavelet shrink methods to remove noise. Experimental resultsshow that small threshold method does well in texture regions, while layeringdenoising method is valid in smooth regions. Finally, wavelet denoising frame based on ROI is built. The method issummarized as follows: If only taking into account interesting regions, there should be three steps:1. Cut the interesting regions off.2. Judgement over the ROI distinguishes texture regions from smooth regions.3. Kemoving noise according to the judgement above.This method above is called ROI cutting filter. If taking the whole image into account, we are specially interested in a givenpart of image, there are four steps:1. Choose ROI and create ROI mask.