Development of Efficient Scalable Video Compression Schemes for H.264/SVC Platform

Abstract

With the rapid development of network technology, the transmission of high quality multimedia data becomes inevitable. In most of the multimedia systems, the clients are becoming heterogeneous on the basis of bandwidth, power and display resolution. Scalable video coding is one of the techniques that supports heterogeneity and encodes a video
stream once, and extracts or decodes in different ways as per the credentials of the receivers. The latest video applications demand video coding techniques that provide various quality levels, spatial resolutions and frame rates encouraging different user preferences, varying transmission bandwidths, and terminal capabilities. Effective adaptation of video content is essential in such applications. Scalable Video coding (SVC) is performed in five major steps; intra prediction, motion estimation, transformation, quantization, and entropy coding at the encoder end and the inverse steps are performed at the decoder end.
This thesis attempts to develop effective scalable video compression schemes suitable for the H.264/SVC platform. It also aims at developing efficacious methods having low
computation cost for low and medium resolution applications by faster motion estimation and compensation. Motion estimation plays a vital role in the overall complexity of the codec, because this is a resource hungry operation for the calculation of motion vectors. The computation time depends on the type of the search technique which is used for the calculation of motion vectors. In this work, an initial attempt is being made for reducing this
computation burden by incorporating a fast and dynamic search method using spatiotemporal neighboring motion vectors, thereby resulting in a faster search to get the best matching block. The performance of this method remains almost on par with the full search technique.
The conventional fast motion estimation algorithms adopt a monotonic error surface for faster computations. But these search techniques may get trapped into local minima during
motion estimation. The overall speed of the algorithm can be further improved by using some heuristic based approach. To address the issue related to the local minima trapping, a
new meta heuristic based motion estimation technique is developed which utilizes Quantum behaved Particle Swarm Optimization (QPSO). This algorithm lowers the computational
burden with improved estimation accuracy. Initially hexagonal search pattern is being used which speeds up the convergence rate of the algorithm. The overall computational burden is reduced to a considerable extent.
In most of the transform based video compression schemes, the wavelet transform is used. But the wavelet based methods do not eliminate the spatio temporal regularity of the video sequence. So an attempt has been made here to develop scalable video compression scheme using Bandelet Transform. In the Discrete Bandelet Transform (DBT), the directions
are modeled by a three directional vector field, known as structural flow. Regularity is decided by this flow where the data entropy is low. The directions of geometrical regularity are interpreted with a two-dimensional vector, and the approximation of these directions is found with spline functions. The bandelet transform based SVC scheme is found to outperform the wavelet transform based scheme.
To further improve the performance in terms of quality and to remove the spatio-temporal redundancy an attempt has been made to implement a hybrid bandelet based scalable video coder structure with Motion Compensated Temporal Filtering (MCTF). The proposed compression scheme provides four scalable layers with six different quantization levels. The
results of the extensive experiments show that this framework enables better peak signal to noise ratio (PSNR), with a reduction in the bit-rate as compared to the traditional Wavelet based Scalable Video Coding (WSVC) scheme. All the proposed schemes are validated with the bench-mark video data sets in H.264/SVC platform and are also compared with the standard Joint Scalable Video Model (JSVM 9.19.14).