Extracting regions of interest based on phase spectrum and morphological approach

Regions of interest extraction in an image is a useful and challenging problem in image processing and analysis. A new method to extract regions of interest based on phase spectrum and morphological approach is proposed in this paper. Different color feature maps (hue, saturation and intensity) in different scales are selected to analyze the saliency based on feature selection. The phase spectrum of the selected feature map is analyzed and the corresponding saliency map is generated. By segmenting the saliency map using a threshold, a segmented binary image can be got. Then morphological operations are done to improve the binary image. According to the improved binary image, the salient regions can be detected in the input image. The method is shown to be effective and efficient through experiments on many natural images. The experiment results and analysis have been presented in this paper.