Efficient Noise Level Estimation Using Orientational Gradient Statistics

Abstract

noise level is an important parameter to improve the performance of many image processing applications. Natural scenes follow special statistics independent of the image contents. These statistical characteristics change under the effects of distortions. Therefore, they can help image processing algorithms to estimate the noise level of input images. We devised a noise level estimation method based on the statistics of orientational differences between image pixel values and those of their neighbors. The proposed approach outperforms the state-of-the-art especially for higher noise levels. This variance estimation approach is effective for both Gaussian and non-Gaussian additive noises. In addition to the high accuracy, because of not using any normalization or image transformation, the proposed method is quite fast and completely useful for real-time applications.