The Research On The Method Of Video Text Detection Based On LVQ

The location, recognition and extraction of the video text, As an important part ofvideo retrieval technology, have become the hotspot to the researchers. Because of the closerelation of the video text and video content, the video text was used for video indexing,which can understand the video content more quickly and locate the interested area of thevideo. Therefore, this paper focuses on the video text detection method, which be used toretrieve and accurate location of the video text, and ultimately for the content based videoretrieval technology.On the basis of the existing text detection method, this paper proposes the method ofvideo text detection based on LVQ. This method aim to solve the low resolution of videoframes, the complex background, the multiple forms of video image frame problems, theconstraints of video text position and video text problems.This paper mainly studies around the following aspects: Through the study of thefeatures and feature extraction method chose the gray feature, the direction feature, themedian characteristics and kirsch operator as the feature vector. Through weighte thedirection feature, the median characteristics and kirsch operator and let them hasdifferentialler ability and certain more denoising ability. The MLVQ network saw which hasthe least Mahalanobis distance as the winning neuron. Using the MLVQ network to classifythe pixel point of the video frame, which initially locate the video text are, and using themorphological method to process that area. Then use OCR software to recognite thecharacter, the ultimate effect is good. The experimental results show that the MLVQnetwork can solve nonconvex and convex region not connected class cannot correctclassification problem, and has higher classification accuracy rate and faster networkconvergence.