Apparatus and method for adaptively reducing noise in a noisy input image signal

Abstract

The basic configuration of Single local Adaptive Window Spatial Noise Reducer (SAW-SNR) is based on a preliminary de-noising low-pass filter followed by homogenous region segmentation to the considered pixel in a given local window. The configuration is composed also of an adaptive local mean estimator, an adaptive local statistic estimator which is preferably an economic standard deviation (SD) estimator and finally, a minimum-mean-square-error (MMSE) based de-noising technique. The proposed segmentation configuration outperforms existing spatial noise reducers in term of subjective and objective performances, in term of edge preservation, noise reduction in both homogenous regions or picture edges and Peak Signal to noise Ratio (PSNR). A second configuration in the form of a Parallel Multiple local Adaptive Window Spatial Noise Reducer (Parallel M-AW-SNR), is a combination of several basic configurations which implements different segmented windows. The M-AW-SNR, which is the less complex configuration for multiple spatial noise reducers, reduces further residual noise as compared to the basic configuration. A third configuration combines the basic configuration of SAW-SNR with a controllable noise variance estimator. This generic configuration allows an adaptive local control of noise reduction level, which can be useful for some correlated noise such as ringing noise in DCT-based decompressed images or cross-luminance noise in composite decoded images.