Scalable Video Compression Framework With Adaptive Orientational Multiresolution Transform and Nonuniform Directional Filterbank Design

What is it about?

Digital video is used in more and more application areas, ranging from Internet video streaming, high-definition TV broadcasting and video telephony nowadays. The demand for high-quality delivery of digital video to devices with broad variety of screen sizes, computing capabilities and network bandwidth conditions motivates the need for scalable video coding (SVC) technology that allows encoding the original video only once and removing parts of the bit stream in order to adapt it to the needs of end users as well as to the capabilities of the receiving devices or the network bandwidth. In order to achieve this goal, image transforms play a key role in this technology as they compress large number of correlated image pixels to small number of independent coefficients. In this research, we invent a novel image transform method based on non-uniform directional filter bank (NUDFB) design and implemented it into the latest-generation SVC framework. This method produces higher compression ratio and allows the decoded video streaming to have better visual quality than the existing other SVC frameworks under the same network condition. We represent the NUDFB design with a non-symmetric binary tree (NSBT) structure in which each depth refers to a decomposition scale and each branch refers to a two-dimensional directional filter bank. Comparing with the non-directional filter banks, directional filter banks can model the images which are full of curves with much less number of transform coefficients and thus improve the compression ratio. Moreover, realistic video contents have different curve orientation distribution but the traditional image transforms do not take it into account. On the contrary, our method analyzes the curve orientation distribution and then provides an adaptive transform by constructing an optimal NSBT structure in which each branch generates nearly equivalent amount of transform coefficients. By this means, the curve orientations that are more frequently observed in the video frames are allocated more coefficients for encoding and therefore more of their information are well preserved. Since our invented novel image transform method is fully compatible to any modern SVC framework, we then implement it into the state-of-the-art SVC framework and compared how our method affects the video coding performance. The current SVC framework does not consider the heterogeneous distribution of curve orientation in video sequences and wastes many bits on encoding them, while our method captures the intrinsic geometric information of the video signal with adaptively designed filter banks and encodes them in much less number of bits. Therefore, the decoded video frames with our method exhibit better visual quality.

Featured Image

Why is it important?

This work is particularly valuable for digital video compression and delivery, especially through the Internet and mobile networks. It provides an adaptive method to compress video frames and reduces the possibility of seeing blurred region. It also saves bits for encoding complex images in the video as more information can be encoded in the same number of bits than the previous image transform methods.

Perspectives