Saliency based Attention Mechanism for Topographic Architectures

Radványi, Mihály and Karacs, Kristóf (2016) Saliency based Attention Mechanism for Topographic Architectures. In: 15th International Workshop on Cellular Nanoscale Networks and their Applications, August 23-25, 2016, Dresden, Germany.

Text Attention_CNNA2016.pdf - Published Version Restricted to Registered users only Download (418kB)

Abstract

Finding outstanding parts in a visual scene is a natural task that humans perform casually using skills learned in early childhood. The process is mostly unconscious, but may also be guided deliberately. This paper presents a model that formalizes abstract quantitative bottom-up cues influencing visual attention. We propose an interestingness metric based on which regions and pattern groups can be selected for processing in order of importance. To show that the cues match well the operator set of cellular visual processors, a sample implementation on the Toshiba-Teli Smart Photo Sensor and detailed results are presented.

Actions (login required)

Edit Item