Object Region Extraction and Visual Attention
Visual Attention has been proved to be useful for object recognition by identifying object region(s) as input to object recognition system (CVPR04, CVIU05). Moreover, selective attention machenism is also useful for other vision tasks as a cueing method. It can reduce the false alarms and improve the performance. But traditional computational attention model is space based. The saliency map only provides 2-D coordinates of salient point, but no information about the region extent of this point. Actually, extracting salient region is the fundamental step for many subsequent processes but not to be a trivial problem. Thus two questions come out from my mind:
1. Can bottom-up selective attention convey region scale (size) information without any priori knowledge of object?
2. If yes, how to calculate this region extent?
From the view of mine, visual attention calculates the contrast to find the salient location (mostly at the boundary of object) and perceptual grouping calculates the similarity to group elements (mostly inside the object). So intuitively, iterative combination of this two process will give both location and scale information of each visual attention. What else can I do? Is it reasonable?
1. Can bottom-up selective attention convey region scale (size) information without any priori knowledge of object?
2. If yes, how to calculate this region extent?
From the view of mine, visual attention calculates the contrast to find the salient location (mostly at the boundary of object) and perceptual grouping calculates the similarity to group elements (mostly inside the object). So intuitively, iterative combination of this two process will give both location and scale information of each visual attention. What else can I do? Is it reasonable?
