IMAGE DENOISING LEVERAGING STRUCTURAL REPETITION VIA LOCAL PATTERN MATCHING

Digital images are often corrupted by noise during acquisition or transmission, which can degrade their visual quality and hinder subsequent analysis [3, 5]. Image denoising is a fundamental task in image processing aimed at suppressing noise while preserving important image features [3, 5]. While numerous denoising techniques exist, effectively handling noise in images containing repeated sub-structures, such as those acquired through electron microscopy of biological samples [1, 2] or manufactured materials with periodic patterns, remains an area of active research. These repeated patterns offer a rich source of redundant information that can be exploited for noise reduction. This paper proposes a method for image denoising that leverages the presence of repeated sub-structures through local pattern matching. The core idea is to identify similar patches or blocks within the noisy image and utilize the information from these redundant patterns to estimate the true pixel values, thereby reducing noise. We outline a conceptual framework involving patch extraction, similarity matching using metrics like normalized cross-correlation, and collaborative filtering or averaging of similar patches in a transform domain or spatial domain. This approach is particularly relevant for images where traditional denoising methods might blur or distort the repeated structures. By exploiting the inherent redundancy, the proposed method aims to achieve effective noise suppression while preserving the fidelity of the underlying patterns.