Energy-efficient approximate 2D Gaussian smoothing filter for error tolerant applications

Present multimedia applications demand highly energy-efficient devices due to need of intensive computation. Most filtering applications engage 2D gaussian smooth filter which is slow and severely affects the performance. In this paper, we propose a novel energy efficient approximate 2D gaussian smoothing filter. The proposed approach is based on “Mirror Short Pixel Approximation” and rounding-off gaussian kernel coefficients. In Mirror Short Pixel Approximation the elements of input image block is replaced via mirror pixel value. The proposed approach is modelled in high level language to check efficacy on benchmark image, which results in minor degradation of image quality. The proposed design is realized on Virtex 6 FPGA. The Simulation results show reduction of 85%, 8%, 58% in area, power and delay respectively compared to existing approximate 2D gaussian smooth filter.