A Non-Local Means Filter with Translating Invariant Shearlet Feature Descriptors

Non-Local computation model focuses on only the complex geometry structure of image, but ignores the appearance model of image. In this paper, we firstly propose a new translating invariant (TI) shearlet feature descriptors which are employed to model the appearance of image patches and we will prove that more accurate measurement of similarity will be achieved using the proposed descriptors. For another, we combined the proposed TI shearlet feature descriptors with non-local computation model to construct a new Non-Local Means (NLM) Filter, which shows great advantage on modeling appearance and structure of image over the standard NLM Filter. Experiment results show that the proposed NLM algorithm gets better performance on natural image denoising than the standard NLM algorithm.