A novel statistical textural distinctiveness approach for robustly detecting salient regions in natural images is proposed. Rotational-invariant neighborhood-based textural representations are extracted and used to learn a set of representative texture atoms for defining a sparse texture model for the image. Based on the learnt sparse texture model, a weighted graphical model is constructed to characterize the statistical textural distinctiveness between all representative texture atom pairs. Finally, the saliency of each pixel in the image is computed based on the probability of occurrence of the representative texture atoms, their respective statistical textural distinctiveness based on the constructed graphical model, and general visual attentive constraints. Experimental results using a public natural image dataset and a variety of performance evaluation metrics show that the proposed approach provides interesting and promising results when compared to existing saliency detection methods.
Illustration of the architecture for salient region detection based on sparse texture modeling and statistical textural distinctiveness: Based on textural representations, we learn a sparse texture model and build a graphical model to computed the final saliency map.
From left to right: Salient objects with texture patterns that are visually significantly different from those of the rest of the scene. Pixels associated with the corresponding atom of a learned texture model. Computed saliency map and ground truth mask.
Code
If you are interested in our software for saliency computation based on statistical textural distinctiveness, we would be happy to share demo code (MATLAB) for research purposes only. You may request your demo code by sending an email to: cscharfenberger@uwaterloo.ca
Downloads
stattexsal_cvpr_scharfenberger.pdf
Related people
Directors
Students
Alumni
Related research areas
Image Segmentation/Classification
Related publications
Conference Papers