Abstract: In order to extract salient object in natural image more precisely, a now method to extract salient object is proposed. First, a fast dictionary learning algorithm is developed for feature extraction based on sparse coding theory. Then, the rarity of the dictionary atoms is computed by the statistical analysis of the corresponding sparse coefficients, and saliency computation is conducted based on the feature rarity. Finally, the mathematical morphology operators are applied to remove the false object. Experimental results show that the proposed method is more accurate for extracting the salient object in natural images than four existing methods and can deal with images containing multiple objects as well.

Key words: information processing technology, object extraction, visual attention, fast dictionary learning, feature rarity, receive operating characteristic(ROC) curve