Abstract
We present a simple yet efficient scalable scheme for wavelet-based video coders, able to provide on-demand spatial, temporal, and SNR scalability, and fully compatible with the still-image coding standard JPEG2000. Whereas hybrid video coders must undergo significant changes in order to support scalability, our coder only requires a specific wavelet filter for temporal analysis, as well as an adapted bit allocation procedure based on models of rate-distortion curves. Our study shows that scalably encoded sequences have the same or almost the same quality than nonscalably encoded ones, without a significant increase in complexity. A full compatibility with Motion JPEG2000, which tends to be a serious candidate for the compression of high-definition video sequences, is ensured.
Highlights
The current video coding standards, such as MPEG-4 part 10 or H.264 [1,2,3], are very good at compressing today’s video sequences at relatively low resolution (QCIF, CIF, or even SD formats)
Video coders based on wavelet transforms (WT) may prove to be much more efficient for encoding high-definition television (HDTV) or digital cinema (DC) sequences
Motion JPEG2000, which extends JPEG2000 to video coding applications, proved to be as efficient as H.264/AVC in intramode for high-resolution sequences encoded at high-bit rate [4] and might be adopted as a future standard for digital cinema and high-definition television
Summary
The current video coding standards, such as MPEG-4 part 10 or H.264 [1,2,3], are very good at compressing today’s video sequences at relatively low resolution (QCIF, CIF, or even SD formats). The input video data firstly undergo a temporal analysis based on the motion-compensated lifted wavelet transform. The delta-filter in the UMCTF framework is a pure subsampling (like in our scheme) and the choice between the deltafilter and the averaging filter depends on the video motion content This adaptation allows a better representation of fast motion in the case of very low frame rate. In our scheme the lowpass filter is not adaptively chosen, since we assume that when temporal scalability is requested, the reference video sequence is the pure subsampling of the original one. A high precision is required in order to obtain regular, differentiable curves For these reasons, model-based algorithms are desirable as long as they can combine computational efficiency with robustness and accuracy.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
