Superpixel Segmentation And Multiple Salient Objects Detection Algorithm With Its Application

Salient object detection is an automatic computer vision process to extract the multiple objects in an input image.However,it is a new challenge since there is much difference of objects size,location and number.Superpixels segmentation is a preprocess for image tasks to enhance the algorithm performance.This paper deals with superpixels segmentation based on the gradient descent way,and designs multiple salient objects detection algorithms for single image based on bottom-up method.Main works and innovations can be summarized into the following three parts:1.Superpixels Segmentation Model based on Nonstationarity Measure: Based on the nonstationarity measure which comes from signal processing field,this model builds the distance measure in the 5D feature space consists of 2D Euclidean space and3 D LAB color space,then improves the clustering result by automatic cluster machine learning,until convergence.Experimental results on the public benchmark database shows the better performance of our proposed approach compared with several stateof-art approaches.2.Mutiple Salient Objects Detection Model based on Adaptive Location: This model starts from the salient points detection and clusters them by AP method,then calculates the number,radius and center of objects to produce an gaussian weight map.To better highlight the salient objects,we adopt the background priors under the framework of geodesic filtering.Experimental results on the two public benchmark databases demonstrate the better performance of our proposed approach compared with several state-of-art approaches.3.Saliency Detection Applied in the Image Cropping: We apply the multiple objects detection maps in the field of contour detection and objects recognition,and do experiment on public database for the image cropping task.We implement the function to automatically crop multiple objects in an image.