Improving feature preservation in high efficiency video coding standard

This paper presents an algorithm to improve feature preservation in High Efficiency Video Coding Standard (HEVC). First, feature information (Keypoints) is extracted from each frame by using scale-invariant feature transform (SIFT) before encoding those frames. Then Largest Coding Units (LCUs) in each frame are categorized into two groups such as important LCUs group and non-important LCUs group based on the number of features in each LCU. After that target bit budget is divided into two parts, one for important LCUs group and another for non-important LCUs group. These two groups are encoded with different quantization parameters (QPs) where QPs values are computed by using their target bits. The experimental results show that our proposed algorithm can better preserve SIFT feature compared with the default parameter in HEVC Video Coding Standard when same bitrate is used.

I. Introduction

According to International Data Corporation report [1], the amount of surveillance video has been exponential increasing and will reach 5800 exabyte in 2020. To reduce the burden of network traffic load, High Efficiency Video Coding (HEVC) [2] is developed as the latest video coding standard which can reduce half amount of the bitrate than its predecessor, H.264/AVC [3], at the same visual quality. However, the existing video coding standards exploits human visual system to achieve high compression ratio while maximizing the subjective quality of the compressed content. This approach can be referred to as human centric approach. As the popularity of computer vision applications (eg., face detection, object retrieval, object tracking, location recognition, etc.) is increased recognition friendliness of compressed videos becomes a research problem where main objective is to preserve as much feature as possible after compression.