Saliency Detection Based On Adaboost Ensemble Learning And Sugeno Fuzzy Integral

As a pre-processing procedure in computer vision,saliency detection takes a very important role in many image processing tasks.In this thesis,two kinds of saliency detection methods are proposed from the perspective of super-pixels classification and feature combination.The first method is based on Adaboost Ensemble Learning model in which two kinds of saliency maps which are based on contrast and classification respectively will be generated.Firstly,extract the background visual information of image based on center-around prior and image structure for super-pixels contrast calculation to get saliency map based on contrast.Secondly,KNN is trained as the basic classifier of Adaboost to obtain a classifier which has strong classification ability.Take this classifier to the test sample to get saliency map based on classification.Lastly,the two kinds of saliency maps are combined together as the final saliency map.The other saliency detection method that we proposed to solve the image features combination is based on Sugeno fuzzy integral.First,divide the image foreground and background regions by convex hulls with foreground seeds extracted and clustered by probability of boundary(PB)and DBSCSN.Second,count the color histogram on the three color features of image and calculate the probability that each super-pixel belongs to the foreground with Bayesian as the confidence in Sugeno fuzzy integral.Finally,calculate the fuzzy measure by the ability of each color divides the foreground and background and then combine the three features of image to obtain the saliency map.We compare the two methods proposed in this thesis with other methods on three public standard datasets from the two aspects of the qualitative evaluation and quantitative evaluation.Experimental results show the effectiveness of our methods and can more properly detect salient regions than other several previous methods.