High-performance fractional anisotropic diffusion filter for portable applications

Abstract

Anisotropic diffusion is one of the most effective methods used in image processing. It can be used to eliminate the small textures of an image while preserving its significant edges. In this paper, a new anisotropic diffusion filter is proposed based on a fractional calculus kernel rather than integer kernel to improve the overall performance of the filter. Integer and fractional anisotropic filters are implemented using the Genesys-2 FPGA kit to utilize the efficiency of parallelism in FPGAs. Integer and fractional anisotropic filters are tested against the achievable PSNR value vs the number of iterations. The proposed fractional anisotropic filter has a better PSNR value using a smaller number of iterations, reducing the power and area compared to integer anisotropic filter. The proposed filter can be used in image smoothing, edge detection, image segmentation, image denoising, and cartooning. In addition, the proposed filter reduces the power consumption by 58.2% compared to integer-order filters, which makes the proposed filter suitable for battery-based applications.