Research Of Semantic-Based Analysis And Segmentation Of Operation Videos

With the development of computer technology and network, multimedia information exchange and new application form has been fully integrated into the people’s daily work, life and entertainment, a large amount of videos generated every day. In the field of medical field also has a large number of video for assistance in education, preoperative education, telemedicine. Media data contains the semantic information such as character, scene, object, behavior and rich story. In order to filter, browse and retrieval, non-linear edit the video efficiently, people proposeand develop content analysis and content based video retrieval technology, in order to achieve the content description and application of semantic level. Let the computer according to the subjective feeling and understanding to represent the media content. How to bridge the semantic gap between low-level features and high-level semantic, to manage and access the videos database by semantic concept has become a challenging research topic in multimedia field.In video content analysis process, video feature extraction and description is a crucial step. While the operation video has its unique features, such as the medical staff in operation room dress color is dark green, in operation room patients in addition to the operation part, the basic operation of cloth covering a shadowless lamp, operation process is in the open state. According to these characteristics, in order to give the operation event modeling, we define the medical staff indicator, operation part detector, shadowless lamp switch indicator and the visual features.Hidden Markov model is a statistical analysis model, although the state it cannot be observed directly, but through the observation vector sequence can be observed. So it is very suitable to the video analysis based on semantic content.. In this paper the hidden Markov model is introduced to analyze the video of operation field. It not only considering the similarity between event characteristics, but also consider the timing relationship between semantic events, providing a theoretical basis for improve the accuracy of video content analysis.In this experiment, five videos of operation have been tagged artificially. The probability distribution of each feature is estimated by the samples, and the state transition matrix is estimated through the Baum-Welch algorithm, reconstruct a hidden Markov model. The experiment is performed using a leave-one-out-cross-validation strategy. For each of the samples to be tested using Viterbi algorithm to identify each video unit in second, resulting in a confusion matrix to present the results of the analysis.The experimental results show that, for semantic features selected, discrimination of some events is relatively high, the correct rate of recognition is high. General operation event recognition rate can reach70%, proved that the hidden Markov model for semantic-based analysis of operation video is feasible.