Development Of Quality-aware Video Systems And NMR Spectrum Registration

Abstract

The work introduces a novel concept of Quality-Aware Video (QAV) System and demonstrates its successful implementation. During the course, it develops algorithms for reduced reference video quality assessment and for data hiding. The idea here is to extract quality defining features of the video sequence and embed them in the original video without causing any perceptual changes to obtain QAV. Such a QAV can then be exposed to distortions and adverse attacks that affect the perceptual quality of the video. At the receiving end, the algorithm extracts the quality features of the distorted video, decode the original video quality features, and estimates the current quality. The beauty of QAV is that they carry the original quality features along with them and hence their quality can be assessed anywhere on the fly. The algorithm developed does not assume any prior information about the attacks which means that the quality assessment is independent of the attack and shows that the algorithm has the potentials to generalize for various attacks. Our second work extends the existing idea of Bayesian estimation for registration to higher magnitude differences. This is achieved by employing pyramid (multi-resolution) approach to the existing algorithm. The advantage of pyramids is at the reduced scale we have better alignment of prominent features (peaks) of the spectrum and as we move to higher levels, the finer details are taken care of. The results have demonstrated an improvement in the existing algorithm.