A probabilistic filter for eliminating temporal noise in time-varying image sequences

The probabilistic filter is a general form that accommodates several types of filters to achieve very good filtering results. Several variations of this filter have been implemented and tested. A fuzzy classification based on the local spatio-temporal characteristics of the image sequence is used to design a probabilistic filter. Motion compensation is used in this filter to effectively remove the noise and preserve motion. The probabilistic filter has proven to be very effective in removing temporal noise in video-image sequences, and increasing the signal-to-noise ratio by 1-3.5 dB. It effectively reduces the noise in the stationary and nonstationary areas without blurring the edges or the spatial details of the image. Where motion compensation fails, the filter reconfigures itself to avoid blurring the edges of the moving objects.<>