Recognition of Urdu text from Video Images

Abstract

As we have huge amount of data in forms of videos and images, therefore we need to understand the content information given in videos and images. Videos are basically sequences of images. Therefore, we need a OCR, which get and recognize text from image. There are many mature OCR of English and other developed languages, but quantity and quality of Urdu OCR is still low. In addition, Urdu OCR is more complex than English OCR due to more characters and connected characters. Most of Urdu OCR recognize Urdu text from document image, which has good resolution and a huge contrast between background and foreground. Therefore, we develop a system named as “Recognition of Urdu Text from Video Images”, which recognize Urdu text from video images. In this system we use Deep learning techniques like CNN and RNN. As we know for Deep learning techniques, we need a huge amount of data for training and getting more accurate system, therefore initially we manual labeled video images by Ground Truth Labeling. We use .NET framework (C# language) for interface design, pre-processing and postprocessing and computing accuracy. For actual recognition of image, we use Tensor flow framework with Python language and develop a model, which recognized Urdu text from image. For making efficient system, we use GPU based tensor flow. Computing accuracy of our system, we use edit distance that provide us accuracy of model with respect to characters.