Texture-Based Text Location for Video Indexing

摘要

This paper proposes texture-based text location methods with a neural network (NN) and a Support Vector Machine (SVM). Both a NN and an SVM are employed to train a set of texture discrimination masks for the given texture classes: text region and non-text region. In these two approaches, feature extraction stage is not used as opposed to most traditional text location schemes, and discrimination filters for several environments can be automatically constructed. Comparisons between NN/SVM-based text location methods and a connected component method are presented.