Convolutional neural network based low complexity HEVC intra encoder
- PDF / 1,650,342 Bytes
- 20 Pages / 439.642 x 666.49 pts Page_size
- 1 Downloads / 159 Views
Convolutional neural network based low complexity HEVC intra encoder Zixi Wang1,2 · Fan Li1,2 Received: 12 September 2019 / Revised: 20 May 2020 / Accepted: 15 June 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Video coding is one of the key technologies of visual sensors. As the state-of-art video coding standard, High Efficiency Video Coding (HEVC) achieves a significant high compression ratio for video. However, it also introduces heavy computational complexity, leading to challenges in application of visual sensors. To reduce the complexity of HEVC intra encoder, this paper proposed a one-stage decision method of CU/PU partition and prediction mode for intra coding. First, the potential factors that may related to the corresponding decisions in CU/PU are explored. Based on this, a one-stage decision network (OSDN) structure is specially designed to determine these decisions. Consequently, the complexity of HEVC intra coding can be drastically reduced by avoiding the brute-force search. Then, OSDN is embedded into the HEVC reference software HM 15.0. Thresholds are set to let the encoder switch between OSDN and the original implementation in HEVC to obtain the final decisions. The experimental results show that the proposed method can reduce 73.69% intra encoding time with 0.1673 dB BD-PSNR loss on average. In addition, the trade-off between RD performance degradation and complexity reduction can be controlled by thresholds. Keywords High efficiency video coding · Complexity reduction · Convolutional neural network · Intra prediction
1 Introduction With the development of visual and camera sensors, video services have been widely applied. Therefore, video technology has become a popular research field and related studies Fan Li
[email protected] Zixi Wang [email protected] 1
School of Information and Communications Engineering, Xi’an Jiaotong University, Xi’an 710049, China
2
Guangdong Xi’an Jiaotong University Academy, Foshan 528000, China
Multimedia Tools and Applications
are derived recently years [9, 19]. Compared to scalar sensors, visual sensors generate a huge amount of visual data. It’s a great challenge to deal with such data, since the current storage and transmission capability are still limited [2]. Therefore, video compression is the key technology of visual sensor application. Due to the heavy amount of visual data and the real-time requirement in application, studying a low-complexity video encoder with high compression ratio for visual sensors is significant. As the latest video coding standard, High Efficiency Video Coding (HEVC) can achieve a significantly high compression ratio. Studies have shown that HEVC saves approximately 50% bit-rate over H.264/Advanced Video Coding (AVC), with similar visual quality of reconstructed video [26]. This improvement benefits from that HEVC introduces advanced techniques, including quad-tree coding unit (CU) partition structure, diverse prediction unit (PU) and transform unit (TU) partition str
Data Loading...
