• Felix Fleckenstein
  • 19.09.2016
  • 17.02.2017

Web conferencing and remote desktop applications require the desktop of one computer to be encoded and transmitted over a network, often the internet, to be displayed on a remote screen. Video compression techniques have to be used to achieve the low data rate needed for real time transmission. However, common video codecs like the well-known H.264/AVC and also the latest standards H.265/HEVC and VP9 are primarily optimized for natural video sequences. To this end, they work using the YCbCr color space involving 4:2:0 chrominance sub-sampling. However, this may lead to annoying artifacts with screen content, e.g. at sharp edges or text.

There are different possibilities to avoid or correct these artifacts. However, to do so in an automatic manner, it is first necessary to automatically detect image areas with a high perceived error. Although there are several widespread image quality measures like the Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Measure (SSIM), these measures are generally inappropriate to describe the perceived error in screen content images introduced by the chrominance sub-sampling. Therefore, a more suitable measure has recently been developed.

The task of this Bachelor's thesis is the enhancement of this new measure with regard to automatic classification of disturbing image areas. Special focus should be put on the verification based on a variety of screen content images, where new test images should be created for.