Research On No-reference Video Quality-assessment Model For Streaming Media Based On Human Visual System

Streaming media video quality will be affected by coding/decoding and network transmission. Meanwhile, with popular application of streaming media, users want to access higher video quality and better user experience. And service operators of streaming media urgently need a real-time video quality method to evaluate the level of their service.The conflict between improving accuracy and reducing computational load is the main problem in real-time video quality assessment. In this paper, based on analyzing the factors causing the distortion of streaming video and video subjective quality assessment experiments, we propose a video quality-estimation model for streaming media based on human visual system. Just using the data packet header information and litter video payload information as input, it has a very light computational load. The experimental results indicate that the model has a high accuracy. With the support of channel control protocol, the model also can assess the encrypted streaming video quality. This paper mainly includes the following aspects:Analyzing existing video quality assessment model for streaming media and according to streaming video distortion characteristics, we propose a new framework of a no-reference video quality-assessment model for streaming media.For constructing the video quality-assessment model, we construct a new streaming video subjective quality assessment database.Through statistical analysis, we derive the functional relationship between the video frame complexity and video frame quantity parameter, the number of encoding bits.According to subjective experiments in which videos are just affected by coded distortion and based on human visual spatial and temporal masking characteristics, the function for assessing video coded quality is derived.According to subjective experiment in which videos are affected by coded distortion and packet-loss distortion, and based on human visual system, mathematical modeling is carried out for different packet-loss conditions.Derived from the subjective data, human visual system has different sensitivities to video segments of different quality. Based on it, a new segmentation weighted method for assessing video sequence quality from every frames'quality is proposed.The model is simulated, and the video objective scores from the simulation are analyzed. The results show the model has a high accuracy.