Saliency Detection based on a Sparsity of Segments Model
Avik Hati, Subhasis Chaudhuri, Rajbabu Velmurugan · 2014
In this paper, we propose a region based saliency detection algorithm using a total variation based regularizer. We aim to obtain salient objects that are uniformly highlighted. The use of the regularizer facilitates the removal of textures from the image. This leads to an image that contains piecewise constant gray-valued segments. This texture-free image is sparsely segmented into a small number of regions using the expectation maximization algorithm assuming a Gaussian mixture model. We compute saliency of regions using their intensity and spatial features. The saliency map is then thresholded to obtain the salient regions of the image. Next we employ an image matting technique to extract the exact boundaries of the salient objects from the image. This approach leads to noise-free saliency maps containing uniformly highlighted salient regions. The experimental comparison with existing saliency detection algorithms demonstrates the superiority of the proposed technique.