Visual attention model for computer vision

In order to model some abilities of the human eye, it is very useful to pay more attention to the visual system organization. In this way the focus of attention principle can be directly described from cone processing when acquiring color images. Cones are photoreceptor cells located in the retina and with the highest density in a small area called fovea. These cells are sensitive to well-defined light wavelengths and are at the basis of color perception. But cones are non-uniformly located over the retina. They are distributed in a radial way from the fovea center. That is why it is necessary to define radial sampling according to a focusing point in order to reconstruct images as they are supposed to be captured by the retina. Such a radial sampling enables as well to obtain compressed data with a large ratio. Although compression is achieved with loss, global information is preserved on the whole image (foveated image). In this paper, the hexagonal cell model is selected to achieve eye saccade mimicking because hexagonal cells are the closest in shape to retinal cells. Furthermore, this cell model gives the best results in terms of data preservation. Such a sampling leads to setting up foveated image processing. In this way, image processing is achieved on less data (15 times less) and so performed in a really faster way. Furthermore foveated images are also used to compute a sequence of points of interest. By following this sequence, a system of vision can mimick eye saccades when focusing successively at each point of interest. This process is a part of the whole process modeling visual attention, as it takes into account detection of points of interest.