Research On The Saliency Detection Of RGBD Image

Visual saliency detection,inspired by human visual attention mechanisms,can simulate the human visual system to quickly detect the most attractive areas in image.As an image preprocessing technology,saliency decetion can help people quickly locate the interest area in image,and allocate resources to deal with it preferentially.It has a wide range of applications in the field of machine vision.This thesis focuses on the saliency detection of RGBD images.The main works are as follows:Focusing on the issue that saliency detection algorithm based on boundary priors was hard to get accurate background seeds and difficult in efficiently measuring the boundary connectivity,a saliency detection algorithm based on minimum spanning tree is proposed.Firstly,a minimum spanning tree is constructed on the input image.then,the most distinctive image boundary is removed,and the remaining three boundaries are used as background seeds.next,searching the shortest distance from each pixel to the background seed on the minimum spanning tree.The path generates a distance map,and uses the part of the pixel with the highest salience value as the target seed to update the distance map.Finally,an auxiliary map is calculated to correct the distance map to obtain a color saliency map on the RGB image.Focusing on the issue that the LBE(Local Background Enclosure)algorithm was over dependent on depth imformation and difficult to fully detect the object which has complex structure,a saliency detection algorithm based on the improved LBE feature is proposed.Firstly,a set of segmentation are obtained by multi-level segmentation.Then,the depth saliency map is obtained by computing and merging the LBE features on each level segmentation.Finally,analysis the characteristics of the color saliency map and the depth saliency map respectively.And a adaptive fusion algorithm is proposed to obtain the final saliency map.Experiment the proposed algorithm on the open data set and compare it with the classic visual saliency detection algorithm.Experiments result show that the proposed algorithm has obvious advantages in precision rate,recall rate and F-measure when compared with the comparison algorithms.And the salience map were more closer to the ground turth map than the comparison algorithms.