Research And Implementation Of No-reference Video Quality Assessment Model Based On Stream Aware

With the continuous development of network technology, now network operators are very concerned about the quality of the network video which is transported by the network which the network operator provided. The quality of the network video is affected by delay, jitter, packet loss and so on. And packet loss has a great effect on the quality of the network video. The network video is always transported without any information of the original reference video in real-time. Thus, studying a network video quality evaluation model which is no-reference, real-time and only consider the impact which is caused by packet loss has become a research hot spot.The quality of network video is not only affected by the value of the packet loss rate, but also the visual characteristics of the content of the area whose packet is lost. In this paper, we study the relationship between packet loss, the visual characteristic of the video content and video quality. We propose a no-reference video quality assessment model based on movement intenseness which consider the type of the frame and the motion intenseness of area whose packet are lost. And we propose a no-reference video quality assessment model based on spatial complexity which consider the type of the frame and the spatial complexity of area whose packet are lost. We propose a no-reference video quality assessment model based on motion vector and DCT based on no-reference video quality assessment model based on movement intenseness and no-reference video quality assessment model based on spatial complexity. And we propose a no-reference assessment method for video quality of spatial-temporal which consider the relationship between temporal complexity and spatial complexity in the video and is based on no-reference video quality assessment model based on motion vector and DCT. These models evaluate the quality of video by analyzing the received stream without fully decoded in real-time.In order to obtain the stable video quality in the client, we need to know the next phase video quality in advance. So we propose a video quality prediction method. This method is divided into three phases. First, we use polynomial fitting model to predict the value of packet loss rate. Second, we use auto-regressive integrated moving average model to predict the next stage of the visual characteristics of the video content. Third, we establish a video quality classification model based on SVM. The feature of this model contain packet loss rate and the visual characteristics of the video content. We use SVM to establish this model to predict the next phase video quality.Finally, we design and implement the network video quality evaluation system which evaluate the quality of network video in real-time. And we analyze the experimental results. The video quality evaluation model in this system is the no-reference assessment method for video quality of spatial-temporal. This system can automatic evaluate the video quality in real-time. We verify the effective and efficient of the no-reference assessment method for video quality of spatial-temporal based on the consistency between the experimental results and the subjective evaluation results and the CPU utilization.