Salient region detection with opponent color boosting

Numerous efforts have been made to detect salient regions in images. Mostly luminance-based saliency models are found in the literature, which ignore the important contribution of color in finding the local distinct image features. Methods about color saliency detection in the literature can only give indication of color salient points or derivatives. In this paper, we present a fast method for detecting the distinct color regions. This model is able to give inference of salient object in the accurate-to-contour level. A segmentation task is also performed based on the proposed color saliency model to show its strength in object outline definition. Finally an evaluation with both the eyetracking and the annotation is achieved for analyzing the model in an extensive dataset. The presented model outperforms the previous work in both tests.