Estimating Salience Using Wavelet Domain Entropy
Our eyes are constantly moving from one location in a scene to the next. In a new scene or image the most visually salient locations or objects capture our attention first. Saliency models based in information theory generally assume that the locations in a new scene or image that first grab our attention are those that have the most information.
We plan to develop an information theoretic saliency model that identifies salient locations as the locations that have the most information and in the model information is estimated by calculating the entropy throughout the image. Theoretical models of the statistics of natural scenes in the wavelet domain have been used as solutions to a number of image analysis problems and this project shows how using these models can successfully model fixation.