Gradients-Predict Filter of Multiple-Scale Template

摘要

Noise filtering is an essential part of any image processor, whether the final image is utilized for visual interpretation or for automatic analysis. Arithmetic Mean Filter (AMF) and Standard Median Filter (SM) tend to work well for fixed-valued impulses but poorly for random-valued impulse noise. The authors review the current trend of image filter development. Some papers present new ways to identify corrupted pixels，while others strongly emphasize on suppressing the noise ratio. And some summarized the both and combined to present new complicated scheme. As widely known, Vector Median Filter (VMF) has the disadvantage of replacing too many uncorrupted image pixels. New method form R. H. Chan, Chung-Wa Ho, and M. Nikolova in 2005 is capable of restoring images corrupted by salt-and-pepper noise with extremely high noise ratio, but the calculation is so complex that it is only can be treated as a post-processing image enhancement procedure. In the fact, the modern imaging equipment is good enough and will not produce more than three percent of salt-and-pepper noise. But as a pretreatment of Image Analysis，filter should focus on preservation the original details and simple or fast for the engineering. Our purpose in this paper is to present a simple scheme to preserve uncorrupted, original pixels, but still enables to remove corrupted ones with a good balance the algorithm complexity and the efficiency.