選択的注意モデルの学習情報の記号化によるハイブリッド画像理解の実現

抄録

Pattern processing and symbolic processing are the major methods used in an image understanding task. Conventionally, they are often implemented and handled as independent systems. However, the handling of real images requires a method that incorporates the characteristics of both. But the designing of a method that enables interaction between symbolic processing and pattern processing is not an easy task. In this paper, we propose a method for symbol-pattern mutual transformation, which, through symbolization of the connection knowledge acquired by the Selective Attention Model, lays the foundation for symbol-pattern integration. We demonstrate the model's effectiveness by applying it to an object segmentation problem.