Fast Trilateral Filtering for Video Denoising
Atishna Samantaray, Saumik Bhattacharya · 2020
Video denoising is an important application of multimedia signal processing to enhance the interpretability of a noisy video sequence. Though several spatial and spatio-temporal filtering techniques are proposed in the literature to perform the task, these existing techniques are typically slow and fail to process video data in real-time. In this work, we present a video denoising technique in the form of a novel trilateral filter. The proposed trilateral filter is an extension of spatial bilateral filter considering the input video sequence as a 3-Dimensional spatiotemporal cube. The proposed nonlinear filtering technique smooths an image along with the preservation of edges and requires a lesser number of iterations for the process. This algorithm uses Fourier bases, the span of the input, and a local dynamic range to compute the trilateral filter output of a given input video. The approximation of the Gaussian range kernels is made by using raised cosines in the present work, however, the proposed algorithm supports any continuous kernel as a range kernel for the processing. The computation is relatively fast as compared to naive trilateral filtering and bilateral filtering, as a lesser number of iterations are involved.