$\lambda$-Domain Perceptual Rate Control for 360-Degree Video Compression

Abstract

The 360-degree video is projected to 2-D formats using various projection methods for efficient compression. As a necessary part of general-video compression, rate control is also indispensable for the projected 360-degree video compression. However, the current rate control algorithm has not been optimized for the 360-degree video compression yet. The Coding Tree Unit (CTU) level bit allocation in the rate control algorithm has not taken into consideration the characteristic that various pixels in 2-D formats have different influences on the visual experiences. In this article, we first propose an optimal CTU level weight taking this characteristic into consideration. The CTU level weight is an approximation to the pixel level weight since the smallest granularity of a rate control algorithm is usually CTU. Second, based on the CTU level weight, a weighted CTU level bit allocation algorithm is proposed to achieve better coding performance. The bits of each CTU are assigned that the Lagrange multiplier $\lambda$ of a CTU is inversely proportional to its CTU level weight. This CTU level bit allocation scheme is applied to all the 360-degree video projection formats. Third, we propose a CTU row (CR) level rate control algorithm for the Equi-Rectangle Projection (ERP) format. Different CTUs in the same row in the ERP format are combined into a CR to provide more stable model parameters. The proposed algorithms are implemented in the newest video coding standard High Efficiency Video Coding (HEVC) reference software. The experimental results show that the proposed algorithm is able to achieve much better subjective and objective qualities as well as smaller bitrate errors compared with the state-of-the-art rate control algorithm.