Study On Interested Object Extraction In Video Image Based On Biological Visual Attention Mechanisms

To overcome the current existing problems of poor real-time performance and low accuracy in the interested object extraction, this thesis establishes a biological visual attention model based on the knowledge of biological visual attention mechanisms. The main contributions are as follows:First, a new biological visual attention model is established based on the knowledge of biological visual attention mechanisms. The final visual attention saliency map is created using a lot of pattern features including motion information features and many static image features. The biological visual attention model can be divided into two sub-modules which connect each other by a serial mode, the low level information extraction sub-module and the high level deep filtration sub-module. The main role of the low level information extraction sub-module is to ensure the real-time performance of the interested object extraction. While the main role of the high level deep filtration sub-module is to ensure the accuracy of the interested object extraction. The knowledge of human and frog visual attention mechanisms is used to build the two sub-modules. Experimental results show that the biological visual attention model established in this thesis has the advantages of real-time and high target extraction accuracy during the interested object extraction in the video image sequence.