计算机科学与探索 ›› 2018, Vol. 12 ›› Issue (9): 1465-1474.DOI: 10.3778/j.issn.1673-9418.1805028

• 人工智能与模式识别 • 上一篇    下一篇

光场熵:针对光场编码的客观评价指标

胡舒童1+,郭碧川2,王    剑1   

  1. 1. 清华大学 电子工程系,北京 100084
    2. 清华大学 计算机科学与技术系,北京 100084
  • 出版日期:2018-09-01 发布日期:2018-09-10

Light Field Entropy: New Metric for Light Field Coding Objective Evaluation

HU Shutong1+, GUO Bichuan2, WANG Jian1   

  1. 1. Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
    2. Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China
  • Online:2018-09-01 Published:2018-09-10

摘要: 近期光场的编码方法使用HEVC编码器达到了较好的编码性能,其中大部分使用峰值信噪比(peak signal-to-noise ratio,PSNR)作为客观指标来评价不同的编码方法。然而PSNR与光场图片的主观质量之间的相关性较差,因其未将光场的各种光学畸变纳入考量,并且其只能对离散像素进行度量而无法度量呈现给观看者的具有连续性的渲染光场图。针对此问题,设计了一个新的针对光场编码的客观评价指标模型。新指标模型中,光学畸变被建模为随机分布,然后使用KL散度度量压缩损失导致分布的偏移。最后,对数种光场编码的客观评价实验比较了新指标与PSNR的性能区别;主观测试的结果显示新指标在预测人眼主观感受的相关性上要优于PSNR。

关键词: 图像编码, 图像质量评价, 图像压缩, 光场

Abstract: Recent researches propose several new approaches to light field image coding that achieve significant compression rate using the HEVC (high efficiency video coding) encoder, most of which use PSNR (peak signal-to-noise ratio) as the sole objective metric to evaluate different coding schemes. However, PSNR does not have a good correlation with the subjective quality of light field images, as it does not take the heterogeneous optical distortion into consideration, and it only measures discrete pixels rather than continuously rendered views presented to  observers. In this paper, a new objective metric is proposed to address these issues. Optical distortion is modeled by a probabilistic distribution, and coding loss results in shifting of the distribution, which is then measured by the corresponding Kullback-Leibler divergence. Furthermore, an objective evaluation of several coding schemes is conducted using the metric along with PSNR to provide a detailed comparison; and a subjective evaluation is performed which shows that the proposed metric outperforms PSNR in correlation.

Key words: image coding, image quality assessment, image compression, light field