Noise Level Estimation on Weak-Texture Image Patch with Image Power Spectrum Sparsity

Abstract

Noise level estimation is important to improve the performance of different image-processing algorithms. Among the different noise level estimation methods, a block-based approach is one of the most effective approaches for estimating the noise level. A noise level estimation method based on a weak-texture patch using the image power spectrum sparsity in the frequency domain is proposed in this paper. A weaktexture image patch is first selected according to the value of image power spectrum sparsity. From the selected weak-texture image patch, the noise variance is estimated by selecting the frequency regions where the image frequency parts are not concentrated. It is observed that the proposed noise level estimation method is effective, especially for images with a rich texture. Furthermore, the proposed method provides a shorter computational time than the conventional methods.