| With the development of society, the computer is able to do more jobs instead of human, and artificial intelligence has a remarkable leap. The computer can not only help humans do lots of repetitive and dangerous work, but also be friend with humans in daily life. As important parts of image processing, foreground extraction and skeleton extraction are paid more attention by researchers increasingly. The skeleton is called medial axis or symmetry axis which is an important feature that can maintain the topology of image and an effective form that express shape. And the skeleton can completely describe the object with less information. Therefore, the concept of skeleton is widely applied in pattern recognition, images search and virtual sport. This article does some research on moving foreground extraction of video image and fast skeleton extraction based on the analysis of latest technology both at home and abroad. And the main contents are as follows.After introducing hardware requirement, software algorithm procedure and some basically theoretical knowledge of the system, this paper compares and analyzes the advantages and disadvantages of some common algorithms for foreground extraction, such as temporal difference and optical flow. Then, after introducing theory and application in background model of single Gaussian background model and Gaussian mixture model, this article introduces the classical mean shift algorithm in detail. After the introduction, a method of background modeling is designed based on the mean shift algorithm. And the foreground extracted from the video image by background subtraction is used to update the parameters of the model, the experiment tests the effectiveness. Based on Euclidean distance extraction, combined with the property that skeleton point is the largest value in one direction, by using the method of local comparison, this paper proposed a new and fast method of skeleton extraction based on distance transform and local comparison. The experiment indicates that this algorithm not only overcomes the disadvantage of thinning algorithm which has many kinds, computational complexity and inaccuracy of skeleton, but also solves unconnectedness of skeleton obtained by common distance transform algorithm. |
