Image Denoise FPGA Implementation Using a Moving Average Filter with Contour Detection
Ricardo Jardim, Fernando Morgado‐Dias · 2018
A Moving Average Filter should be among the first filters, if not the first, one should consider when speed, good precision and low to medium hardware resources are what is required for implementing image noise reduction on a physical device. The technique is fairly simple to implement, but if taken a straight forward approach, the result will be a blurry image which is undesirable. Since the human eye is most sensitive to the sharp tone transition, we introduced a contour detection technique. Because the inherently lack of floating point representation, except for the most modern FPGA's versions, all the calculations were done in fixed point. Also, images used were scaled up to 10×103colour levels (a bit higher than 213bits). Real results of double precision 64-bit were achieved through Matlab and then compared to a Zynq-7000 FPGA using 64-bit fixed point calculations. The error difference obtained between both implementations was 1×10−4.