Similar Patch Selection in Embedding Space for Multi-View Image Denoising

Geunwoo Oh1, Dong-Wan Choi2, Bochang Moon1

GIST 1 , Inha University 2

IEEE Access

BM3D denoising results (a) without (top) and with our technique (bottom). We visualize the number of similar patches per pixel given the same user threshold (τ ) for both approaches (b). We set the number of candidate patches for our approach to be the same as the one used in BM3D (e.g., 1521 = 39 × 39). BM3D finds similar patches whose patch-wise l2 distance from the reference patch is less than τ in a spatial window (e.g., 39 × 39). When it fails to find enough numbers of similar patches, the denoising result leaves low-frequency artifacts (see the top row in (c) and (d)). Our technique increases the number of similar patches by feeding a set of candidate patches identified from the entire inputs (multi-view images) to BM3D, and it allows the method to produce an improved result (see the bottom row in (c) and (d)).


This paper proposes an image patch selection that finds similar patches in multiple images so that image denoising can suppress noise more effectively by exploiting the identified similar patches from the multi-view images. We encode all image patches in multi-view images into a low-dimensional space, and it allows for a denoiser to find similar patches effectively from the space. Our approach enables existing patch-based denoisers, which often find similar patches within an image window, to identify more similar patches by extending the limited search space into the entire space (i.e., all input images). We integrate our technique into state-of-the-art single-view denoising (block-matching and 3D filtering (BM3D)), and demonstrate that the BM3D combined with our approach is able to conduct multi-view image denoising effectively, without a major alteration to the existing algorithm.