Research On Image Segmentation Method Based On Co-saliency Detection

Human beings obtain a large amount of image information and their understanding of the outside world through vision.How to separate the target object of interest from the huge image data volume is a hot spot in the field of computer vision in recent years.Therefore,it is particularly important to achieve a significant segmentation of the image.Importantly,the feature selection and accurate segmentation of the image association in the image collection is the main research direction of this article.This paper studies the image segmentation based on collaborative saliency detection and the image segmentation based on deep learning.The main work of the full paper is summarized as follows:(1)An image segmentation algorithm based on patch cluster and multi-measure saliency detection is proposed.The image segmentation algorithm based on patch and multi-measure cluster saliency detection is a simple and accurate saliency detection method.It establishes the association of the images in the image collection,which clusters the image patchs and calculates four kinds of significance measures.The saliency images are fused to get the final segmentation result.Experiments show that the patch cluster and multi-measure saliency detection method can not only improve the segmentation accuracy,but also has high robustness.(2)An image segmentation algorithm based on patch labeling and contour prior information is proposed.The image segmentation algorithm based on patch labeling and contour prior information performs image segmentation by finding the target contour of the image in the image set and marking the region.The image is divided into tree patchs and label the internal and external regions while the edge contour map of the salient target object is obtained.Calculating and merge the saliency images to get the final segmentation result.Experiments have proved that the algorithm is fast and accurate.(3)A segmentation model for plant image collection is proposed.By constructing and labeling their own plant image data sets,the deep learning algorithm framework is used to train the model and segment the plant images.The experimental results show that the model has excellent segmentation effect for plant images and can achieve accurate instance segmentation.