AIDI: An adaptive image denoising FPGA-based IP-core for real-time applications

Abstract

The presence of noise in images can significantly impact the performances of digital image processing and computer vision algorithms. Thus, it should be removed to improve the robustness of the entire processing flow. The noise estimation in an image is also a key factor, since, to be more effective, algorithms and denoising filters should be tuned to the actual level of noise. Moreover, the complexity of these algorithms brings a new challenge in real-time image processing applications, requiring high computing capacity. In this context, hardware acceleration is crucial, and Field Programmable Gate Arrays (FPGAs) best fit the growing demand of computational capabilities. This paper presents an Adaptive Image Denoising IP-core (AIDI) for real-time applications. The core first estimates the level of noise in the input image, then applies an adaptive Gaussian smoothing filter to remove the estimated noise. The filtering parameters are computed on-the-fly, adapting them to the level of noise in the image, and pixel by pixel, to preserve image information (e.g., edges or corners). The FPGA-based architecture is presented, highlighting its improvements w.r.t. a standard static filtering approach.