Abstract
Additive noise removal is an important and practical problem in image processing. Bayesian and other approaches have been proposed to restore the corrupted images. More methods are proposed to preserve signal details while removing noise such as the bilateral filtering. This filter is a weighted average of the local spatial and range distance over a set of data. However, users have to find acceptable two parameters σS and σR in bilateral filters manually by their experience. In this paper, we propose an algorithm which can generate two parameters automatically by estimating standard deviation in noisy images. We also apply this algorithm on images with large additive Gaussian noise. Experimental results compared by PSNR show that our algorithm is better than using manual parameters. We also show that the effect of the domain transform, color noise and window size on the noise removal. We develop a software and hardware based system for this adaptive bilateral filtering. The adaptive parameter estimation algorithm is done in the software and the bilateral filtering is in the hardware. Based on UMC 90 nm process, we design and implement the first real-time bilateral filter that can process 180 frames of size 720x480 per second.