Abstract
We present an active vision system which chooses and fixates the scene with the largest information. We use Shannon information to measure image features, and present an information map to quantify a saliency map. Particularly, we use intensity, hue and saturation as the image features to be quantified. The system also has a task dependent attention selection ability. To achieve a given task, the system selects some features, uses them to calculate the image information, constructs an information map, focuses on the area with the largest information, and zooms in it. Experimental results demonstrate that the system can acquire images from a real scene with the highest information and generate compact images with different resolution.
