Simultaneous Detection And Recognition of Named Entities in Degraded Document Images

Polar Rain proposes to design and develop end to end name entry extraction system for degraded handwritten documents. We will develop a new data and computational model that combines both image features and language model to extract name entities in degraded documents. Our proposed approach brings innovative solution to work on degraded Arabic printed/handwritten documents. We will investigate how to design such data model which contains both image features and language features together, and can handle complex degraded document images. Our computational models will use generative and descriptive data models to extract names from recognition results that may contains significant error. We will design a flexible software architecture so that the solution can be extended to other similar languages, such as Pashto, Dari without extensive work. We will conduct experiments to measure performance of proposed system over exiting NEE solution.