Bottom-Up Visual Attention Guided by Genetic Algorithm Optimization

E.T. Pereira, H.M. Gomes, and V.F.C. Florentino (Brazil)


visual attention, genetic programming, computational grids, image region location.


This paper discusses the application of genetic program ming in the optimization of the weigths of a visual attention mechanism when performing two specific visual tasks: to locate people and object regions in an image. A saliency map, which indicates the most salient regions in an image, is created from the weighted sum of a number of interme diate feature maps (such as Intensity, Color and Orientation at different scales). A genetic algorithm with overlapping populations was employed to find the best set of weights that minimized the number of salient points needed to lo cate a particular class of image region. Experiments with a different set of images have confirmed the improvements in region location using the optimized weights.

Important Links:

Go Back