Integrated Action AT2009-0026

Visual Attention using Combinatorial Pyramids


The aim of this project is to develop an artificial visual attention model. This mechanism will be responsible of extracting the relevant objects of the real scene from the broad input of visual information provided by the acquired video sequence. In order to build the saliency map associated to the sensed scene, a novel perception-based grouping process will be employed. This grouping process will be performed by a hierarchical irregular structure which should be able to preserve the topological information at upper levels. Thus, these levels could be used to perform the perceptual grouping process. Specifically, this project proposes to study the application of combinatorial pyramids to accomplish this task. Taking into account low-level image features (e.g. colour contrast, edge or depth information), the perception-based grouping process will extract from the input video sequence, the different parts or ‘pre-attentive objects’. These parts will correspond to units of visual information that can be bound into coherent and stable objects. In addition, the ability to handle dynamic scenarios will be included in the proposed model by introducing a tracking mechanism for moving objects. This mechanism will be also performed using the same hierarchical structure, reducing the computational time.

Programa de Acciones Integradas
Duracion: 1 Enero 2010 - 31 Diciembre 2011


Last modification: 19.11.2010