Novel Approach for Salient Region Detection

Abstract: Salient region detection in image is useful for applications like adaptive compression, image segmentation and region-based image retrieval. Many of existing saliency detection techniques do not pay much attention to the noise problem. In this paper the proposed model uses L0 smoothing filter to reduce the effect of noise and achieve a better performance. The important role in our framework is played by the L0 smoothing filter. The L0 filter is extremely helpful in characterizing fundamental image constituents, i.e. salient edges and can simultaneously reduce insignificant details, thus producing more accurate boundary information for background merging. The saliency maps in this paper are extracted by three features: the color spatial variance, border measurement, and region contrast. The color spatial variance is utilized to calculate the degree of color distribution. Based on the observation that salient objects are less likely to be connected with the borders of an image, the boundary information of each segment is used to determine the salient region. Last but not least, a region contrast method is used for extracting global contrast used to achieve full resolution map with well defined boundaries. Thus this method will give more accurate results as compared to other methods.

Keywords: Image segmentation, L0 smoothing filter, Region contrast,Sailent region detection