Study And Implementation Of Salient Region Extraction Based On Attention Mechanism

With the development of computer science and technology, the color image and the video applications are used more and more broadly, because the final observer of images is human, it is an important way to study the human-eye vision characteristic and find a method of salient region extraction that can imitates the human-eye vision characteristic. how to extract video salient region effectively has been one of the hot topics of digital image research recently. Up to this time, Researchers have brought forth many methods of salient region extraction which can advance the development of video compression and video retrieve research.This paper mainly introduces attention mechanism, image segmentation and the theory and main methods of extraction salient region, then study the visual salience feature and feature extraction, in the last salient region extraction system based on attention mechanism is introduced.The paper mainly includes:1. This paper introduces the main methods of video salient region extraction proposed in recent years, and introduces human visual attention mechanism into video salient region extraction, then studys and realizes a new extraction technology based on the attention mechanism. Finally, designs and builds video salient region extraction system based on this salient region extraction technology.2. A method of salient region extraction based on attention mechanism is designed. First, the picture is segmented by homogeneous region, then the salience of the homogeneous region is analyzed and salient regions are extracted. Basing on this model, we propose a new method to select salient regions. It suggests that the salient regions extracted by this method tally closely with the regions of human attention in the result of the experiment. It provides an effective content representation model for senior application like video analysis and search, etc.The result of salient region extraction method based on attention mechanism is compared to the result of the ScaleSaliency method in this paper,and the result is analyzed.Finally, sum up the whole paper and point out some problems and extentions left to be further studyed.