Multi-dimensional WFM filter and its hardware realization for impulse noise removal

A Multi-dimensional Weighted Fuzzy Mean (MWFM) filter used in color image filtering is proposed and analyzed in this paper. MWFM is an extension of the Weighted Fuzzy Mean (WFM) filter with embedding of a fuzzy detector and a dynamic selector into WFM to overcome the poor performance of WFM in detail signal preservation. The fuzzy detector uses two fuzzy intervals and refers to the WFM-filtered outputs to detect the amplitude of impulse noise which will be used by the dynamic selector. Its detection capability is better than that of the conventional detector weighted by a crisp interval. Moreover, it results in high stability over the full range of noise occurrence probability. By means of the dynamic selection approach, MWFM not only preserves the high stability and performance of WFM in removing heavy additive impulse noise, but also improves the performance of WFM for light additive impulse noise. Therefore, it can filter corrupted color images well under from noise-free conditions to noise-full conditions. For dedicated hardware implementation, the kernel of the MWFM filter, WFM, is synthesized using generic LR fuzzy cells which realize high-speed fuzzy inference. The hardware complexity is much simpler than that of the conventional median filter, and simulation results show that up to 6.6 M pixels per second can be filtered by a WFM filter with a small chip area.