Evaluation of Denoising Algorithms for Source Camera Linking

Abstract

In the source camera linking (SCL) tasks, a large number of images taken by the same camera is not available, then forensic investigation must be carried out using only residual noises extracted from each natural image without any knowledge about their source camera. Therefore, an efficient denoising algorithm and/or noise enhancement function are required to estimate accurately Sensor Pattern Noise (SPN). In this paper we provide a systematic evaluation of common-used denoising algorithms with different parameters under a SCL task. The denoising algorithms considered are locally adaptive window-based denoising and the Block-matching 3D (BM3D) denoising. The SCL task used for evaluation is image clustering based on their source camera, in which we construct three sets with natural images taken by 5, 10 and 15 different source cameras. Linkage clustering algorithm with Ward modality is applied to group the images by their source camera. The experimental results show that the BM3D denoising with standard \((\sigma )\) deviation of \(5\) provides the best performance. Said method achieved a clustering accuracy of 98%, 96% and 85% for 5, 10 and 15 cameras respectively.KeywordsPhoto Response Non-Uniformity (PRNU)Sensor Pattern Noise (SPN)Source camera linkingDenoisingSPN enhancementHierarchical clustering