Description

We propose a new noise detection technique that makes use of a detector based on new statistics called robust outlyingness ratio (ROR), used to measure how much a pixel looks like an impulse noise. We do not directly use ROR to detect the outliers; we just use it to measure the outlyingness of each pixel in the image. Based on the ROR, the pixels are divided into four different clusters. Then, different decision rules are adopted to detect the impulse noise in each cluster. During the detection process, the from coarse- to-fine strategy is used. Different thresholds are used in the two stages. Finally, the detection procedure is iteratively adopted. We concentrate on universal noise and a universal noise-filtering framework based on the nonlocal means (NL-means). The NL-means is very powerful for denoising Gaussian noise, but to the best of our knowledge, it is not yet extended to remove impulse noise. Our proposal is modeled to improve NLM for both kinds of noise.