Video caption could be used to index video stream with high-level semantics since it implied lots of semantics inherently. The prior work of caption location and extraction considers how to define good caption features and neglects the self-generalization of classifier machine thereof. In order to overcome this limitation

an algorithm firstly localization and extraction video caption using support vector machine (SVM) and independent component analysis (ICA) is presented. In this algorithm

the raw video frame is segmented into N * N sub-blocks

and each block is identified either a caption block or a non-caption block; then mutually high-order independent ICA features are used to train a support vector machine classifier; finally the location and extraction of video caption can be finished automatically with pyramid model and de-noising techniques by each trained support vector machine classifier. Because support vector machine holds excellent generalization of classification with non-enough samples and independent component features are naturally high order independent each other

compared to other algorithms

the experiment data shows this method works well.