Research On Saliency Detection Based Video Attention Complexity Measurement Algorithm

With the development of multimedia information technology,not only can people assess the quality of images and videos in pixel level,for example by analyzing noise and resolution,people can also fulfill the evaluation in semantic level by measuring the content of images and videos.Video attention complexity being a kind of feature extracted from video sequences themselves can include a variety of video content information.Apart from objectively describing the video content,these information can,to some extent,reflect the subjective sensation of video from people.Since video attention complexity can play an important role in many practical applications,obtaining the method of measuring video attention complexity should be a research object of vital importance.In this paper,we propose a algorithm to measure video attention complexity.Firstly in this algorithm,bottom-up and top-down features of video frames are considered to get saliency maps for every frame of videos.Secondly,a saccadic model based on saliency maps,several oculomotor biases as well as foveation mechanism will be used to predict gaze fixation positions for videos.Thirdly,motion estimation based salient objects analysis also performs in the framework in order to get the final score of video attention complexity.Compared to other related algorithms,our proposal not only considers low level visual features,but also takes high level features into account.What's more,the characteristics of human visual system and human viewing habits modeled in our algorithm could offer better simulation of human observation behavior.With the contributions mentioned,this algorithm could help obtain video attention complexity scores which are more close to actual situation.In this paper,video attention complexity scores obtained from our algorithm will be compared to subjective experiment scores and other related measurement algorithms.Experiment results show that our algorithm has its feasibility and the scores obtained are consistent with subjective experiment scores.Moreover,our algorithm shows certain advantages in the comparison with other related algorithms.