06095070

a

06095070 Dozal, Le\'on Olague, Gustavo Clemente, Eddie S\'anchez, Marco Evolving visual attention programs through EVO features. Di Chio, Cecilia (ed.) et al., Applications of evolutionary computation. EvoApplications 2012: EvoCOMNET, EvoCOMPLEX, EvoFIN, EvoGAMES, EvoHOT, EvoIASP, EvoNUM, EvoPAR, EvoRISK, EvoSTIM, and EvoSTOC, M\'alaga, Spain, April 11--13, 2012. Proceedings. Berlin: Springer (ISBN 978-3-642-29177-7/pbk). Lecture Notes in Computer Science 7248, 326-335 (2012). 2012 Berlin: Springer EN evolutionary visual attention organic genetic program evolved visual operators evolved feature integration artificial dorsal stream

doi:10.1007/978-3-642-29178-4_33

Summary: Brain informatics (BI) is a field of interdisciplinary study covering topics such as attention, memory, language, computation, learning and creativity, just to say a few. The BI is responsible for studying the mechanisms of human information processing. The dorsal stream, or ``where''stream, is intimately related to the processing of visual attention. This paper proposes to evolve VAPs that learn to attend a given object within a scene. Visual attention is usually divided in two stages: feature acquisition and feature integration. In both phases there are specialized operators in the acquisition of a specific feature, called EVOs, and on the fusion of these features, called EFI. In previous research, those referred operators were established without considering the goal to be achieved. Instead of using established operators the idea is to learn and optimize them for the visual attention task. During the experiments we used a standard database of images for visual attention. The results provided in this paper show that our approach achieves impressive performance in the problem of focus visual attention over complex objects in challenging real world images on the first try.