We propose a rotationally invariant similarity measure as a modification of the well-known block matching algorithm for finding similar regions in an image or an image sequence. This algorithm can find similar patches even if they appear in several rotated or even mirrored instances. We demonstrate the application of this approach to enhance the quality of the non-local means algorithm for image denoising. For this filtering method, we also introduce a locally adaptive way of choosing the parameters. Numerical examples show that both modifications lead to a visible and measurable qualitative improvement of the denoising results.
展开▼