Toward Semantic Visual Attention Models J Kucerova, Z Haladova |
|---|
Comenius University in Bratislava, Slovakia
|
| Visual information is very important in human perception of surrounding world. In the visual perception of the environment, specific parts of the observed scene are salient, i.e. more important than others. Visual attention is the ability of a visual system to detect these salient regions in the observed scene. In our work, we are focusing on detection of these salient regions in a complex scene using visual attention models. We utilize the visual attention model, which is based on local context suppression of multiple cues [Hu et al, 2005, Proceedings of IEEE ICME 2005, 346-349]. The model implement 3 attention cues: color, intensity and texture. We have extended this model with the semantic information about the scene by creating a new visual cue, the map of the occurrence of the important object. The importance of an object is usually very individual and task dependent. However some objects proved to be salient generally (faces, text..). Our approach can be utilized for the extraction of the salient regions in both task based (find object X in scene) and general situations. The information about salient regions in a scene can be further used in image compression, thumbnailing or retargeting. |
| Up Home |
|---|