Image Denoising Based On Statistical Model And Non-local Means Algorithm

Due to the imperfection of image acquisition systems and transmission channels, images are often corrupted by noise. This degradation makes more difficult to perform high-level vision tasks such as recognition, 3-D reconstruction, or scene interpretation.First, combining with Bayes least squares estimator, we describe a method for removing noise from digital images, based on GSM with Gaussian–Hermite PDF in Steerable pyramid domain in this paper, which can be seen a modified version of the BLS-GSM. By introducing the Gaussian–Hermite PDF, we model the distribution of Steerable pyramid coefficients with GSM. The statistical model, and then used to obtain the quantization noise image decomposition coefficient through bayesian least squares estimate. Experimental results show that the method can be widely used for better performance than the subjective and objective evaluation in the most advanced denoising technology.Second, the nonlocal (NL) means filter is a typical technique for denoising textured images.However, this algorithm only defines up to translation ,without considering the direction and scale of the image patch. In order to improve the image quality, proposed Pseudo-Zernike moments based on non-local means algorithm for image denoising. Pseudo-Zernike moments is based on the image of the entire region shape description operator, not only has rotation invariance and stronger ability to resist noise ,but also can struct any higher moments , therefore using the Pseudo-Zernike moments into NL-means filter can get much more pixels or patches with higher similarity measure. The comparative experimental results show that the improved NL-means filter achieves higher peak-signal-to-noise ratio and obtained better denoising performance.