12704_SCTE_Broadband_Nov2016_COMPLETE_lowres

technical

Online video will be responsible for four-fifths of global internet traffic by 2019.

video quality differently. The visual system is not the same for each person, and our sensitivity level to different aspects of the video is not identical. Moreover, the understanding of how human beings perceive visual information is not complete. This situation is further complicated by the multi-dimensional complexity that is involved in video scenes. This article provides a comprehensive overview of the history of objective video quality measurement, touching on the current methods used as well as new research into the HVS that enables operators to deliver more consistent, superior video quality. Common methods of video quality measurement For many years, Picture Signal-to-Noise Ratio (PSNR) has been the most common method to measure video quality, despite the fact that many other new and improved methods are available. PSNR is a method that mathematically calculates the distortion of each pixel in the image. This method completely ignores the impact of the surrounding pixels, the complete image and the video scene information. We will discuss, later in this article, the impact of the surrounding pixels on the measured distortion as it appears to the HVS.

in different scenes. For example, PSNR overestimates distortion when the surrounding area is spatially active and underestimates distortion when the surrounding area has low spatial activity. (See Figure 1 below.) In Figure 1, images (a) and (b) have the same PSNR score. 2 However, PSNR underestimates distortion in low spatial activity areas; it calculates low distortion for the dotted area in the sky, which is actually a very visible distortion. In 2004, Structural Similarity (SSIM) was introduced, offering a better video quality measurement than PSNR. A key element of the HVS is to identify shapes and structures in the image. The HVS is less sensitive to the details of image pixels and more sensitive to the shapes of the objects and details of the image. Leveraging its higher sensitivity to structure, which in turn relates to shapes, SSIM is able to achieve a more accurate video quality measurement compared with PSNR. SSIM independently measures structural and non-structural distortion. While differently weighting the structural and non- structural distortion, the correct balance can be achieved. SSIM takes into account three elements of distortion: average luminance, average contrast and structure/texture similarity. Mathematically, the latter two components combined are

Since PSNR does not take into account any masking effects imposed by the surrounding pixels, distortion is miscalculated

Figure 1: PSNR

2 Figure 1 reference: Video Quality Assessment Methods: A Bird’s-Eye View I P. M. Arun Kumar, S. Chandramathi

45

Vol. 38 No. 4 - November 2016 Issue