Abstract: By combining bottom-up and top-down information, a novel visual attention model based on blob-guided perceptual grouping is proposed in this article. The model can build knowledge representations for prior information by means of blob features through introducing multi-level blobs and connecting blob properties and low-level features. For any new given scene, the model can use the prior knowledge to render the object features more salient by enhancing those characteristic features of the object, and then it groups regions together recursively to form objects, guided by blob feature vectors extracted from the intermediate data at the pre-attention stage. Selective visual attention in the model can be effectively directed to task-relevant regions. A comparison of the model with other attention models which can direct attention to salient proto-objects proves its superiority.

Key words: image processing, vision, remote sensing, feature extraction, object accumulation