Video Text Detection Based On Multi-feature Fusion

In recent years,with the continuous progress of the country and the rapid development of science and technology,the number of videos has been growing at lightning speed.Video is widely used in the fields of education and teaching,aerospace,intelligent transportation,biomedicine fields.And video plays a very significant role in the way for humans to obtain information.In the past,simple keyword search technology has been unable to satisfy human query for massive video data.People want to query in such a huge video database to find video information that they want or interest,which is becoming increasingly difficult.Compared to the underlying information such as color,edge,etc.,the text in the video contains a lot of useful content and can be a good overview of video information.The emergence and rise of Support Vector Machine(SVM,Support Vector Machine)has attracted the attention of many scholars and put forward the method of using SVM to realize the detection of video text.For these methods,single features,excessive feature dimensions,long time consuming,and unsatisfactory detection results.That is,it can not take into account the deficiencies of time complexity and detection effect,this paper made an in-depth research and improvement,and proposed two new methods of video text detection based on SVM to improve the detection effect of text.Finally,through the analysis and comparison of the results,it is proved that the improved algorithm presented in this paper is indeed effective in achieving the effect.This article first summarizes and analyzes the relevant background and research significance of video text retrieval technology,and this paper analyzes and summarizes the research status of video text detection by scholars at home and abroad.Based on the study of related theoretical knowledge and feature extraction methods,two algorithms for detecting video texts in combination with SVM are proposed.The specific research work is as follows:(1)This paper proposes a method for detecting video text by extracting the integrated features of text color,edge and texture and combining SVM.Firstly,we need to intercept a certain number of positive and negative samples from the video,which contains the non text and text area.Then the selected samples are extracted based on color,texture and edge features.Then according to the characteristic data of the positive and negative samples obtained to train and get the corresponding SVM classification model.Finally,the corresponding video text detection is performed according to the obtained classification model.Finally,according to the experimental results and data comparison,the method proposed in this paper can achieve a good improvement in the detection results.(2)At present,there are many documents using the HOG and SVM method for pedestrian detection,and achieved good results.Based on this,the text attempts to extract the HOG feature of the text,and combines color and edge features,after training the characteristics,the text will be tested according to the training model and finally get the text area in the video.In the same way,the steps of sample selection,feature extraction,model training,and text detection are performed.Finally,by analyzing the experimental results and the evaluation criteria of the control algorithm,it is proved that the algorithm can effectively improve the accuracy of text detection and reduce the false detection rate.(3)Most of the detection of video text based on existing literature is based on a single frame.When the dynamic background is more complicated and the margin is stronger,the false detection rate is higher.Therefore,this paper analyzes the shortcomings of detection based on single frame,and improves it based on the detection of adjacent three frames on the basis of the algorithm,so as to reduce the false detection rate of the text,thereby improving its detection efficiency.