Research On Computational Model Of Visual Attention

Attention selection is regarded as an essential cognitive mechanism of human visual system, Which relies on visual attention mechanism that selecting the relevant parts of scene and simultaneously ignoring subordinate information, Since information from only a small region of the visual field can progress through the cortical visual hierarchy, visual task can be effectively and securely dealt with by limited processing resources.Computer vision system faces the same problem with humans:there is a large amount of information to be processed and to achieve this effectively even satisfy real-time application, the order in which a scene is investigated must be determined in an intelligent way, A promising approach is to use computational attention system to simulate human sensory.This dissertation addresses the research on visual attention mechanism and its computational methods, the neurobiological mechanism of visual attention and spatio-temporal saliency detection model are analyzed, then a multi-feature integration model based on neurophysiology theories of human visual saliency is proposed, then a new approach for visual attention selection called Multi-Channel Amplitude Modulation Fourier Transform is established which is based on two up-to-date works:Spectral Residual model and Phase Fourier Transform, this method adjust image amplitude spectrum for energy balance resolving the problem that low-filter salient object was inhibited with current frequency-domain saliency detection algorism. In addition, in terms of extraction of focused region, we substitute channel tracing region growing algorithm of color image segmentation of for traditional ones in synthesized saliency map which trade-off the feature according to their contribution to saliency, so that the extracted result corresponds to human sensor and achieve better simulation experiment effect.This dissertation exhaustively represents theories and approach of computational model of visual attention selection and proposes two types of visual attention model which has a strong theoretical and applied value. Additionally, this paper proposes a series of practically valued evaluation standard in the view of imperfect of evaluation method of attention selection model and verified feasibility and validity of our model with this evaluation standard.