A low complexity texture-discriminating noise removal method for video encoding

A classification-based, low-complexity motion-compensated spatio-temporal noise reduction filter that preprocess a video sequence before encoding is presented in this paper. The classification of small blocks of each frame into texture, edge and homogenous are applied before filtering. And then motion-compensated temporal and spatio-temporal filtering are applied to classified small blocks accordingly to get better filtering results while trying to preserve edges and details as much as possible. The complex motion information finding process of motion compensation in filtering process is avoided by using the readily available motion information from encoding of previous frame so that computational complexity of the whole filtering can be greatly reduced. The proposed algorithm can significantly improve the visual quality of noisy video sequences by maintaining sharp edges and some details, and clearly smoothed out the noise from most parts in the processing time of about 40 frames per second. Experimental results show that it can improve coding efficiency of JM software in terms of the bit-rate reduction between encoded filtered and noisy video sequences while preserving the video quality efficiently.