ON BITSTREAM FEATURES BASED PERCEPTUAL QUALITY ESTIMATION OF HEVC CODED VIDEOS

Abstract: ABSTRACT Context: High Efficiency Video Coding (HEVC) is a new video coding standard which combines high video quality with higher compression ratios. In order to fully use the potential of this standard, there should be created an appropriate coding tool (coder), which determines proper coding parameters to ensure the highest possible video quality while maintaining a specified bitrate. Objectives: This thesis aims at proposing a set of bitstream based features that can be used for perceptual quality estimation of HEVC encoded videos. To this purpose, we develop required computer programs capable of extracting these features from coded video files. The extracted features are used in an artificial neural network (ANN) based model to estimate video quality. Methods: To conduct our solution, we performed a profound analysis of the HEVC coding standard, and then we designed software that precisely retrieves all needed data from video files. The software was created in C# language in order to allow for the analysis of big sized XML files. The other programs were created in Matlab software. They contain file converters and ANN video quality predictor which perform prediction of quality values on the basis of extracted parameters values. Tests were performed on 560 sample video files. Results: ANN provided very good results in quality prediction. All experiments showed that for all tested quality metrics presents very good fit. The highest correlation coefficient R is for VIFP quality metric and is averagely equal to R= 0,992. Conclusions: Summarizing presented approach for extraction video parameters and quality prediction can be implemented in real experiments. Provided testing conditions allow to achieve for satisfying results.