Reduced complexity post-processing of block-based coded video sequences

In a video communication chain, the limiting factor to the nal received images quality is the limited bandwidth of the delivery medium and the displaying system. Today, the advanced video compression techniques combined with the appearance of high de nition LCD/Plasma screens providing higher native resolutions than the traditional CRT monitors, led to considerable improvement of the decoded picture quality. However, video compression standards induce several artifacts into the decoded video sequences which are divided into two major categories: spatial impairments, including the blocking and the ringing artifacts and temporal impairments including the mosquito noise and the background area ickering. The objective of this thesis is to propose reel time solutions in order to reduce these compression impairments. Concerning the reduction of the spatial impairments, we proposed two algorithms, the rst one dealing with the blocking artifacts and the second one with the ringing artifacts. The proposed deblocking lter is mainly based on an adaptive conditional two-dimensional lter, derived from the combination in horizontal and vertical directions of a simple two-mode conditional 1-D lter. Appropriate lters are selected using the local degradation of the image, which is assessed by examining the quantization step as well as the computed spatial pixel activities. The ringing artifact reduction algorithm uses a simple classi cation method to di erentiate at and edge blocks, which are then ltered using a particular weighted median lter. Regarding the temporal impairments, we proposed a new measure to assess the level of these impairments and accordingly estimate the temporal quality of the decoded sequences. The preliminary study of the temporal compression artifacts demonstrated that the level of the temporal uctuation between consecutive frames is a ected by the compression ratio and the group of pictures structure, notably the presence (and period) of intra coded frames in the video sequence. It was also shown that the deringing process reduces the visibility of the mosquito noise, however a temporal ltering remains necessary to reduce the background areas uctuation. For this reason, we proposed to temporally lter these areas and skip the temporal ltering of moving objects. The implementation complexity of the proposed solutions was investigated in order to prove the applicability of these solutions for real time applications. This complexity was assessed on a 3 GHz processor as well on a 350 MHz media processor. Overall, this thesis has demonstrated that the proposed solutions can be used for rapid processing and e ective reduction of three main annoying compression artifacts, namely blocking, ringing, and temporal ickering.