A Novel No-reference Video Smoothness Evaluation Method Based On ORB Descriptors

With the popularization of computer and maturing of network technology, the network multimedia applications associated with videos have got rapid development. Video quality assessment has become an indispensable part of multimedia system. In terms of the video a benchmark sequence quality evaluation, this paper proposes an evaluation method of no-reference mobile video smoothness.There are a variety of video image quality evaluation methods, usually from aspects such as clarity, color cast, contrast and signal-to-noise ratio to measure, but detection of image clarity is not only difficult but also focus. In this chapter, we mainly do from the aspects of definition to study two kinds of typical video quality problems and detection and assessment. The first category is the mosaic problem, it is a typical result in loss of information due to the quality of the transmission network disruption caused, and adds the mosaic of artificial factors. The second category is the blot problem. It is generated by the excessive number of digital playback media player or contamination. This chapter also describes the general characteristics of image feature extraction and the method of extracting features of classification.The final acceptor of video is the human eye, so the human eye is the most accurate judge of the video quality. The video jitter(i.e. video smoothness) is another important indicator of the video quality and is directly related to the people’s perceptions of the video. Video smoothness is one of the important indicators to evaluate the quality of a video. In terms of the video a benchmark sequence quality evaluation, this paper proposes an evaluation method of no-reference mobile video smoothness. No-reference method to be evaluated based only on video, can be completely out of the reference video quality assessment for the target video. First, we describe the video frames in a sequence by extracting their ORB descriptors. Next, we calculate the translational and rotational displacements between two adjacent video frames by matching their respective ORB key points using the FLANN(Fast Library for Approximate Nearest Neighbors) algorithm. This is followed by the acquisition of the translational and rotational trajectories of the video sequence by using a two-dimensional motion model. For diminishing the random jitter present in the trajectories, Gaussian filter is applied. We propose a novel smoothness evaluation metric which is based on the translational and rotational jitter frequencies of the trajectories. The experimental results show that the proposed algorithm can cater for the uncertainties encountered by the existing methods, accurately detect the jitter in the video and hence realize the no-reference evaluation of video smoothness.