Correcting temporal artifacts in compressive video sampling with motion estimation

Abstract

We present an algorithm for finding solutions to combat artifacts due to temporal coding of video sequence. The compressive sensing method has been utilized tremendously in recent years to solve underdetermined problems. The use of compressive sampling in video coding is promising since video signal has a sparse property and its volume is massive. In order to reduce spatial and temporal redundancy, we apply DCT and some motion estimation algorithms. Then we integrate those methods with the projection transformation based on compressive sampling principles. Despite the large amount of data volume in a video, we can still meet the requirement on sparsity in the residual frames resulted from motion estimation and compensation process. The basic motion estimation, i.e. the differential method, cannot handle the temporal artifacts reliably, especially for video sequences with fast motion objects. However, we have to consider the constraints of computational complexity at the encoder side. In this research, we compare the reconstruction results of various block matching algorithms, such as differential method, three step search, new TSS, and the simple and efficient search.