摘要

A new framework is presented for saliency based object detection in naturally complex scenes. By identifying the ground-object structure through the scene images, in this framework, the random distribution of chromatic information is articulated into saliency patterns with respect to a scene specific primary system. Due to the intrinsic coherence of the scale and chromatic randomness, detected saliency patterns are well localized as object images within the perspective induced by the ground-object structure. By using the saliency pattern, the computational complexity of landmark detection can be significantly reduced.

  • 出版日期2012-3