Research On Interactive Video Segmentation Based On Enhanced Onecut

Video segmentation is a classification problem,which aims to separate the object from the background.Video segmentation is an important part of video semantic analysis,and also an essential part of the machine vision and artificial intelligence,video segmentation is widely used.Onecut model is a kind of Graphcut model,which is widely used in image segmentation.Since the flexibility of the energy function,the Graph cut model has good ability of feature constraint and fusion.Therefore,the energy function model has been widely used to solve the problem of image and video segmentation.Since a lot of video segmentation methods take advantage of the spatial and temporal continuity between frames,but consider the time and space equivalent when calculating the inter frame relations.In this paper,the Onecut model is enhanced by combining the global probability model and the local classifiers model and the temporal and spatial constraints,then the Onecut is successfully applied to video segmentation.Firstly,this paper presents an interactive video segmentation method based on color feature,shape feature and motion information of object.After the interaction On the key frames is added,the accurate contour of the object is obtained by the traditional interactive image segmentation method.Establish the global and the partial classifier.The estimated initial contour According to the contour,which is transferred from the previous frame outline to the current frame by patch matching,is abtained to update the parameter of classifiers on current frame.Then the enhanced Onecut model is used to obtain the accurate segmentation results.Secondly,this paper also takes into account the difference in the dimension of time and space,and enhance Onecut segmentation model based on the optical flow and bilateral space constraint.The original Onecut model and enhanced Onecut based on classifiers was compared with two kinds of temporal and spatial constraints which is optical flow and bilateral space.Compared with other video segmentation algorithms the DAVIS,our enhanced Onecut algorithm proposed in this paper has higher segmentation accuracy and improve the spatial and temporal continuity.