Saliency is the property of an object to “stand-out” with respect to its surroundings. The goal of saliency detection algorithms for images is to assign a value to each pixel indicating how distinctive or contrasting it is with respect to its pixel neighborhood.
|(a) Original Image||(b) Saliency Map|
An object, in this case the yellow flower, can stand out with respect to its surroundings, the purple flowers and the floor, in terms of color, texture, shape, and various other features. Our goal is to generate full-resolution saliency maps that indicate salient objects in pixel precision.
Generally speaking, saliency detection algorithms differ in terms of the features that are used, and the size of the pixel neighborhood that is considered, for computing the distinctiveness. At IVRL, we have developed several state-of-the-art saliency detection algorithms and applications for them.
- FASA: Fast, Accurate, and Size-Aware Salient Object Detection (ACCV 2014)
- Saliency Detection Using Regression Trees on Hierarchical Image Segments (ICIP 2014)
- Saliency Detection Using Maximum Symmetric Surround (ICIP 2010)
- Image Summaries Using Database Saliency (SIGGRAPH Asia 2009)
- Frequency-tuned Salient Region Detection (CVPR 2009)
- Salient Region Detection and Segmentation (ICVS 2008)