Study Of The Sudden Visual Target Capture And Recognition Based On Visual Attention Mechanism

The human brain has unparalleled information processing capacity in the face ofcomplex scenes, however the computer vision often appears to be inadequate in theprocessing of visual tasks under the conditions of structured environment. Scientists tryto explore the human eye mechanism which will be modeling and simulating oncomputer for natural image understanding and object recognition. This visualmechanism is selective attention mechanism that makes the eyes focus on region ofinterest (ROI) to obtain the accurate identification of the scene target information in thehuman world and has a strong adaptability and stability. In recent years, this attentionmechanism is widely used in computer visual to reduce the calculation amount ofprocessing information and to improve the computation efficiency of visual processingalgorithm. This paper will do research in the recognition of the sudden visual targetbased on Visual Attention Mechanism (VAM). The major works are listed as follow:1. An algorithm of the sudden visual target fast recognition is proposed by usingboth VAM to generate static saliency map and optical flow method to generate dynamicsaliency map, so as to obtain so called combined static and dynamic the total saliencymap which is done by using linear weighted, then, the ROI is detected, optical flowmethod is used to track and recognize.2. In consider of the computational complexity and a great deal of time-consumingto do image global optical flow entire calculation, an improved method, so calledoptical flow method which will do optical flow tracking on some reliable points ofcharacter in the ROI, is developed, and then scale change for non-rigid target motion arecompensated for warping function.3. Finally, three algorithms, which are optical flow method, Mean shift algorithmand Kalman Filter, are independly used in two videos to do processing, analysis andcomparison. The experiment results show that the optical flow method is moreadvantageous than Mean shift algorithm and Kalman Filter in the robust tracking andaccurate identification of the visual targets.