摘要
为了降低新一代高效视频编码(HEVC)的计算复杂度,融合基于人眼恰可察觉失真(JND)的快速帧间模式选择和自适应编码单元深度范围(ACUDR)两种算法,提出了一种快速帧间编码新方法。首先,通过建立时空域JND和最佳预测模式的统计关系,利用JND阈值指导帧间预测模式的快速选择,以减少遍历的模式个数从而降低复杂度;然后,利用编码的前一帧定义与当前帧的平均深度差(ADD),通过ADD自适应加权时空域相邻最大CU(LCU)的平均深度值预测当前LCU的深度范围(DR)类型,减少当前LCU遍历的深度个数从而降低算法复杂度;最后,融合这两种算法以进一步降低了视频编码的计算复杂度。结果表明,所提出方法在低延时和随机访问两种编码结构下,降低了40%以上的计算复杂度,但编码性能降低很小,在同等码率下的峰值信噪比(BDPSNR)仅分别降低0.058dB和0.037dB,同等客观质量下的码率(BDBR)仅分别增加1.71%和1.06%。
In order to reduce the computational complexity of the emerging high efficiency video coding （HEVC）, a novel fast inter coding method is proposed, which combines just noticeable difference （JND）- based fast inter mode selection and adaptive coding unit （CU） depth range （DR） decision algorithm. First of all, for reducing computational complexity, we decrease Candidates for predicting mode through a JND threshold which is obtained by building correlation between the best prediction mode and spatia~ temporal JND. Thus, depth range （DR） type of the current largest coding unit （LCU） is estimated through the weighted value of spatial-temporal adjacent LCUsr average depth with average depth differ- ence （ADD） which is defined by current and previous frames. Thus, the unnecessary depth from candi- dates can be reduced. Furthermore, the above two algorithms are fused for further reducing computation- al complexity of HEVC. Experimental results show that in the low-delay and random-access cases,the proposed algorithm can save encoding time of more than 40 ~ with only Bj. ntegaard delta peak signal- to-noise rate （BDPSNR） losses of 0. 058 dB and 0. 037 dB,Bj, ntegaard delta bit rate （BDBR） losses of 1.71 % and 1.06 %, respectively. The proposed method also saves the same encoding time compared with Xiong＇s method with a BDPSNR gain of 0. 019 dB and a BDBR gain of 0. 5% in low-delay case.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2015年第5期932-941,共10页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(U1301257
61171163
61271270
61271021
61311140262)资助项目
