Automatic Subjective Quality Estimation of 3D Stereoscopic Videos: NR-RR Approach

A method for estimating subjective quality score of 3D stereoscopic video is proposed which is based on decision trees. The output of this estimation can be fed into encoding and transmission units for compensation. The proposed method operates with minimum dependency on reference video. Content characteristics, no reference (NR) and reduced reference (RR) quality metrics are extracted and summarised prior to training stage. Content features are based on spatio-temporal activities within depth layers. Quality features include NR blockiness, NR blurriness and RR 3D stereoscopic video quality metric. Due to fast and accurate requirements for the quality estimation, decision trees are employed where a 0.94 accuracy is achieved.