Segmentation of occluded objects using a hybrid of selective attention and symbolic knowledge
説明
Key fields in image understanding are pattern processing and symbol processing. Often, these processes have been studied independently. In this paper, we propose a hybrid system for early stage image understanding in which symbol processing and pattern processing are used complementarily and in parallel. Here, the common knowledge that is necessary for their interaction is automatically acquired using the image recollection ability of the Selective Attention Model. Local pattern features are converted into a symbolic description. We have developed an image segmentation system which is able to recover the lacking part of given object using symbolic inference. The system was applied to an occluded image and the usefulness of the hybrid system was demonstrated.
