Saliency motivated improved simplified PCNN model for object segmentation

Abstract

Abstract Based on the fact that PCNN and saliency detection method all can achieve the better simulation of HVS to locate the objects that have the most interests in an image, a novel approach for object segmentation, termed as saliency motivated improved simplified pulse coupled neural network (SM-ISPCNN) algorithm, is proposed in this paper. Instead of adopting pure gray-scale to activate the ISPCNN neurons, it is better to introduce the saliency feature value to motivate this model. The introduced saliency stimulus applies a sliding window to precisely exploit the distributions of the objects and surroundings, weakens the influence of background while retaining the region of interest; the highlight of this ASLFC saliency feature lies on: (1) the saliency estimation is based on semi-local areas instead of pixel level; (2) the estimations of the conditional distributions is manipulated via integral histogram approach; (3) the adaptive prior probability setting method is employed to achieve more promising saliency map. For improvement of convergence speed of SM-ISPCNN model for object segmentation, at each iteration, we regard top 5 regions as feedback input for next iteration, which can raise the robustness of SM-ISPCNN model against noise and other interferences. We demonstrate the proposed model based on the mammograms from open and common database of MIAS, gray images from Weizmann segmentation evaluation database and color images from public database with ground truth annotations. Compared with five competitive methods, our model has the obvious superiority for segmentation capacity and algorithm robustness, furthermore, it does not requires any training and can be used in various occasions of object segmentation. In addition, this method is verified on the mammograms from Gansu Provincial Cancer Hospital, the detection results reveal that this model has great potential in clinical application.