VLSI Implementation and Comparative Analysis of Anisotropic Diffusion Filters PMAD and FORADF for Real-Time Applications

Abstract

 In the era of miniaturization, designing an energy-efficient, high-quality image processing system for portable devices is a challenging task. Image processing and VLSI design are the two domains that can contribute toward achieving this goal. Images are often cor
rupted because of various types of noise during the different stages of image processing, from image acquisition to display. One of the widely occurring noises is the salt and pepper noise, which can be removed using median filtering. Median filtering preserves the edges, but it does not provide any smoothing effect. The visually appealing smoothing effect is very much essential in digital cameras nowadays. Anisotropic diffusion filters are well known for their smoothing effect with edge preservation. The foundational aniso
tropic diffusion filter, Perona & Malik anisotropic diffusion (PMAD), can remove Gaussian noise with edge preservation, but it failed under the perturbation of the salt and pepper noise. According to the literature, many anisotropic filters developed over decades for the removal of the salt and pepper noise from PMAD resulted in significantly higher hardware complexity compared to PMAD. The first-order robust anisotropic diffusion filter (FORADF) is well-suited for removing high-density salt and pepper noise with edge preservation and smoothing properties. The hardware complexity of FORADF is also much lower than PMAD. In this paper, the hardware implementation of PMAD and FORADF is discussed. RTL netlist, schematic view, and synthesis reports of both filters are generated using Xilinx Vivado 2023.1 and the Genus tool of Cadence. The comparison of the filters is conducted quantitatively and qualitatively. The comparative analysis undertaken in this paper shows that FORADF is well-suited for low-power, real-time applications.