A Learning Based Approach to Validate Text in Video
一种基于学习的视频字幕验证方法

For improving accuracy,validating text is a key step of detecting text in video.The current approaches mostly based on experiential rules.The approaches are not adaptive,in condition of complex background,low resolution,varied font,size,color of text in video.For improving adaptability and accuracy of validating text,the application of two-dimension principal component analysis(2DPCA) for video frame processing is investigated and a novel 2DPCA and support vector machine(SVM) based approach for validating text in video is proposed.The approach has two steps of training and validating.Firstly,2DPCA is adopted to get the features of video image patches.Then,SVM is trained to validate and classify video image patches.The experimental results illustrate that the novel approach for validating text in video is more effective and costs less time than the other approaches,in condition of complex background,low resolution,varied font,size,color of text in video.