Efficient frame complexity estimation and application to G.1070 vide quality monitoring

ITU has standardized a computational model as Recommendation G.1070 for Quality of Experience (QoE) planning [1]. In our previous work, we proposed a system for calculating the G.1070 visual quality estimate in a monitoring scenario [2]. In G.1070, the visual quality is based, in part, on frame rate, bitrate, and packet-loss rate. For a fixed frame rate and a fixed packet-loss rate, the G.1070 visual quality score will decrease with decreases in bitrate. However, G.1070 can not distinguish between cases in which a decrease in bitrate truly does represent a decrease in quality and cases in which the underlying content is easy to encode, thus resulting in a lower bitrate without a corresponding decrease in quality. In this paper, we propose a modification to G. 1070 model to account for this difference by including an analysis of the underlying complexity of the video content. More specifically, we propose a quality measure in which the bitrate input to G.1070 is replaced with a normalized bitrate, where the normalization is based on an estimate of the complexity of the compressed content. With this proposed enhancement to the model (named as G.1070E), it allows a much better approximation to MOS values and the NTIA-VQM[3].