Simultaneous segmentation and recognition of Farsi/Latin printed texts with MLP

A new segmentation and recognition method for multi-font, multi-size Farsi/Arabic text using Multilayer feedforward neural networks is presented. The automatic feedback between segmentation and recognition completely removes or at least remarkably reduces the amount of errors due to segmentation. Furthermore, the common failures in recognizing Farsi/Arabic printed characters and in detecting some special lines are reduced. The OCR recognition method uses MLP type neural networks with Fourier descriptors.