面向数字驱动近眼显示的位平面运动估计算法

Bit-plane Motion Estimation for Digitally Driven Near-eye Display

近眼显示器在高分辨率和高刷新率的前提下应尽量小型化,传统视频压缩方案难以满足需求。将人眼临界频率与偏心率的关系量化,提出视点恰可察失真模型用于位平面运动估计的匹配准则。将运动估计的搜索范围优化为时间维度和灰阶维度两部分。结合人类视觉系统和概率统计分析,新增补充匹配块替代残差数据。制定基于位平面运动估计的视频压缩方案以适配数字驱动近眼显示器,以现场可编程逻辑门阵列为核心设计控制器并搭建系统进行验证。实验结果表明,对低五个位平面压缩效果最均衡,压缩比为1.385,数据传输量恒定,有利于硬件设计。峰值信噪比为37.658 dB,结构相似度为0.975,还原后的图像与原图像无明显差异,符合人眼直观感受。

Near-eye display should be miniaturized under the premise of high resolution and high refresh rate,traditional video compression schemes can not meet the demand.The relationship between critical frequency of human eye and eccentricity is quantified,and a viewpoint just noticeable difference model for matching criteria of bit-plane motion estimation is proposed.Search range of the motion estimation is optimized into two parts:time dimension and gray scale dimension.Combined with the human visual system and probability statistical analysis,supplementary matching blocks are added to replace the residual data.A video compression scheme based on bit-plane motion estimation is developed for digitally driven near-eye displays,a controller is designed with field programmable gate array as the core and a system is built for verification.The experimental results show that the compression effect on the lower five bit-planes is the most balanced,the compression ratio is 1.385,and the data transmission volume is constant,which is beneficial to hardware design.The peak signal to noise ratio is 37.658 dB,and the structural similarity is 0.975.There is no obvious difference between the restored image and the original image,which is in line with the intuitive perception of the human eye.