The Multi-object Saliency Detection Method Based On Deep Feature

With the rapid development of internet technology,the visual saliency detection has gained more and more achievements in image processing,and widely applied in many fields,such as image compression,scene recognition,target tracking and location,etc.However,the existing methods mostly rely on the low-level features defined by human and are only for the image with single saliency object and relatively simple backgrounds,which greatly limits the efficiency and accuracy of the saliency detection for the relatively complex situations.In recent years,deep learning provides a new idea in image saliency detection.According to the images including multiple saliency objects with the relatively complex background information,a multi-objective saliency detection method and application based on the deep feature are studied in this paper.According to the problem in expressing complex image of low-level feature,a deep feature extracting method through the deep convolutional neural networks is proposed.Firstly,the input image is divided into superpixels in several scale spaces,and the prior knowledge is used to extract and optimize of preselected object regions in the meantime.Then,the local and global features of preselected object regions are extracted with the deep convolution neural network,and the principal component analysis is used to reduce the deep feature dimension to calculate the saliency value based on different feature.Because of the saliency calculation based on the superpixels,the saliency values in the same saliency object are inconsistent and will affect the overall extraction of saliency objects.In order to solve this problem,a weighted improved multi-layer cellular automata is used to optimize the multi-scale segmentation saliency maps to get the final saliency map.The effectiveness of the proposed method is verified by experiments on standard datasets SED2 and HKU_IS,comparing with typical saliency detection methods.Finally,according to the saliency detection result of the single image,a multiple images co-saliency detection method based on the deep feature matching is designed and applied to the video saliency objects detection and co-localization system.