Research On Interactive Video Object Extraction Based On Gradient-Constrained SLIC Algorithm

Video object extraction is the basic but important technology in multimedia data processing,using the user's mark on the video frame which is used as prior information of the video object,extracting the video object in all video frames and ensuring that different video The relevance of the video target on time and space.Extracted video objects can be commonly used in post-production,3D modeling,intelligent monitoring and other fields.However,the current video object extraction methods cannot get ideal extract result in the natural video fragments with complex content(video containing video objects and background areas,video target boundary blur,illumination changes and shadows,etc.),which means there is a unsatisfactory situation(like incomplete video object)often in the extract result.On the other hand,the user's request for the efficiency of video object extraction is getting higher,and the current video object extraction methods are less efficient.Aiming at the problem of extraction quality and extraction efficiency in video object extraction method,this paper focuses on the two aspects:video preprocessing and feature measurement mechanism,to research the above problems.We propose a fast video object extraction method based on gradient constraint SLIC method,using less user interaction on the key frames,this method can quickly and-accurately extract the video object from the videofragments.The algorithm this paper proposed is mainly divided into following three steps:Firstly,in video preprocessing,this paper first analyzes the existing video preprocessing methods,and finds the limitation of these methods.Then,this paper adopts the improved preprocessing method to preprocess all the video frames,which divides each video frame into several independent and disconnected sub-patches.Secondly,in the construction of three-dimensional undirected graphs,this paper fully measure the correlation between sub-patches in spatio-temporal domain,and constructs the three-dimensional undirected graph with the sub-patches,while building the corresponding energy function.Finally,in the aspect of feature measurement,this paper redefine a new robust measurement distance function by considering Appearance information(color and texture)and motion information,to achieve high-quality video object extraction for complex video fragments.On the other hand,to ensure the spatio-temporal consistency of the video objects in the extraction results,this paper adopts high-order potential term in energy function,and improves the quality of extraction result by adjusting the proportion of motion feature in the global feature measure function.The experiments show that our method can achieve high quality video object extraction even dealing with high-definition video fragments with complex scenes,and the time efficiency of video object extraction is obviously improved compared with the existing video object extraction method.