Recognising text in real scenes

摘要: We present two different approaches to the location and recovery of text in images real scenes. The techniques we describe are invariant scale 3D orientation text, allow cluttered first approach uses page edges other rectangular boundaries around locate a surface containing recover fronto-parallel view. This is performed using line detection, perceptual grouping, comparison potential regions confidence measure. second low-level texture measures with neural network classifier an image. Then view each located paragraph by separating individual lines determining vanishing points plane. illustrate our results number images.