In a method of and an apparatus for processing image data representing an image saliency data for the image is generated by determining a series of features of the image and the determined features are used to generate a probability measure for each point of the image representative of a location of a subject of the image. The saliency data is processed using respective ones of weighting functions of a plurality of spatial scales in order to determine the positions of regions of interest of the image at respective ones of the scales. Response data is generated for each scale representing the relative strength of response of the saliency data at the positions of the determined regions of interest to the function at that scale.