Ming-Ming Cheng 1 Guo-Xin Zhang 1 Niloy J. Mitra 2 Xiaolei Huang 3 Shi-Min Hu 1 1 TNList, Tsinghua University, 2 KAUST/IIT Delhi 3 Lehigh University Reliable estimation of visual saliency is an important step in many computer vision tasks including image segmentation, object recognition, and adaptive compression. We propose a regional contrast based saliency extraction algorithm, which simultaneously evaluates global contrast differences and spatial coherence. Our algorithm consistently outperformed existing saliency detection methods, when evaluated using one of the largest publicly available data sets. We also demonstrate how the extracted saliency map can be used to create high quality segmentation masks for subsequent image processing. Abstract Sample results Evaluation using the largest public available dataset (containing 1000 images) Histogram Based Contrast (HC) Region Based Contrast (RC) Application: Content Aware Image Resizing Application: Saliency Cut Global Contrast based Salient Region Detection Histogram based speed up Color space smoothing 1 ( ) ( , ) n l j l j j Sc fDcc 2 ( , ) ( , ) ( ) exp () s k i s k i D r r r r k k r i i S D r r r r Segmentation 2 0.4 s 2 s Region contrast by sparse histogram comparison. Spatial weighting Application: Non-Photorealistic Rendering http://cg.cs.tsinghua.edu.cn/people/~cmm/ free! Input IT MZ GB SR AC CA FT LC HC RC RCC Recall Precision Input Saliency maps Saliency cut Region size Limitations Examples Our Our RC based saliency Cut achieves P = 90%, R = 90%, compared to previous best results P = 75%, R = 83% on this dataset. Our Our HC RC