Region of interest extraction based on visual attention model and image clustering analysis

In view of the Itti' visual attention model not searching the shape and edge of the region of interest, a novel Bottom-Up visual attention model based on traditional Itti' model is proposed which includes wavelet for the image down-sampling and local iterative method based on 2-D DoG filter for the feature combination strategy. The visual saliency map is generated by the modified visual attention model, combining with the color image clustering analysis result to realize the automatic ROI extraction. The experiment result shows a better extraction effect than traditional visual attention model.