Multi-Fractal Articulation of Environmental Saliency Arising in Naturally Complex Scenes

Abstract

A new framework is presented for saliency based object detection in naturally complex scenes. In this framework, the random distribution of the scale and chromatic aspects of image complexity are exploited for the multi-fractal articulation of scene image into the ground-object structure and object images, respectively. Due to the intrinsic coherence of the scale and chromatic aspects, detected object images are well organized within the ground-object structure to indicate various types of maneuvering context arising in the scenes.