Text line detection and localization is a crucial step for full page documentanalysis, but still suffers from heterogeneity of real life documents. In thispaper, we present a new approach for full page text recognition. Localizationof the text lines is based on regressions with Fully Convolutional NeuralNetworks and Multidimensional Long Short-Term Memory as contextual layers. Inorder to increase the efficiency of this localization method, only the positionof the left side of the text lines are predicted. The text recognizer is thenin charge of predicting the end of the text to recognize. This method has showngood results for full page text recognition on the highly heterogeneous Maurdordataset.
展开▼