Detecting Perceptually Important Regions in an Image Based on Human Visual Attention Characteristic

Abstract

In this paper a new method of automatically detecting perceptually important regions in an image is described. The method uses bottom-up components of human visual attention, and includes the following three components: i) several feature maps known to influence human visual attention, which are computed in parallel directly from the original input image, ii) importance maps, each of which has the measure of “perceptual importance” of local regions of pixels in each corresponding feature map, and are computed based on lateral inhibition scheme, iii) single saliency map, integrated across multiple importance maps based on a simple iterative non-linear mechanism which uses statistical information and local competence of pixels in importance maps. The performance of the system was evaluated over some synthetic and complex real images. Experimental results indicate that our method correlates well with human perception of visually important regions.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.