Research And Application Of Compute Model On Visual Attention Mechanism

At present the study on the combining of computer science ,biophysics and neural physics is hotspot, and the Essential Question is visual attention model. Visual attention mechanism is the unique human physiological characteristics. It plays an important role between humans and external environment of understanding. If we use the visual attention mechanism on processing with the machine visual,it will bring huge benefits for human work and life. The resarch on Visual attention calculation model can help people screening analysis of large amounts of data, enhance the computer processing data information ability, it has an important application value in the image analysis , image understanding field, target detection and machine vision .This paper first introduced the physiological basis of human visual, elaborated Itti visual attention calculation model of key technology and experiment, through the studying on the Itti calculation model found that the model of image of significant goals have strong recognition ability, but in image understanding of the contents exist certain deviation, including some results and human cognitive interpretation, there exist discrepant pseudo focus, leakage detection and attention focus unreasonable, etc.The paper has made some improvements on the Itti model to improve the capacity of image recognition. Firstly the paper made some improvements on the method of significant and use the global evolution to count up all the significations ,then combine both of the results to generate the significant point merge image and find out all the regions of interesting . The method can maintain the more important information features than Itti model on the ability of descriping the whole image. The experiments show the method can find the goal exactly and accord with human cognitive level better.Finally the paper uses the method of improving computation model on the compressing images. Fistly it finds out the interested regions in the image, next use JPEG2000 algorithms on the interested regions and background regions. The method not only greatly reduces the image size, but also guarantees the good regions of the image. The result has been proved to be success.