Saliency Detection Based on the Bayesian Model of Improved Convex Hull

Due to the limited precision, we improve the saliency detection of Bayesian model in this paper. First, the improved convex hull is got by using the intersection about convex hull of Harris corner of the original image and compressed image. Second, we get the prior probability map by combining with space subspace clustering and improve convex hull. Then, observation likelihood probability is computed by using the color histogram. Finally, we compute the saliency map according to the existing prior probability map and observation likelihood probability by using the Bayesian model. And the finally saliency map is optimized. The experiments are tested on data sets of MSRA and SED. Experimental results showed that our algorithm not only preserves the vision effect, but also improves the performance evaluation of precision-recall curves, F-measure and MAE values.