Pre-Processing of Document Images Obtained with a Smartphone
Hassan El Bahi, Abdelkarim Zatni · International Review on Computers and Software (IRECOS) · 2016
In recent years, text preprocessing has become increasingly important in the field of pattern recognition because of its use in various domains. The greatest difficulty is to build an effective text preprocessing system able to overcome the problems of perspective distortion, nonuniform illumination and poor focusing. Many systems are being proposed, but less interest has been given to document images acquired with a smartphone camera. In this paper, a complete text preprocessing system of document images obtained via mobile phones will be presented. The system comprises three steps: initially, a method is proposed based on edge detection, morphology operation and heuristic rules to extract the text area from document image. In the second stage, an approach was developed to cope with the problem of perspective distortion by using a new method that relies on polar coordinates and bilinear interpretation. After that, a simple method based on projection profile was proposed in order to eliminate the marginal noise. Finally, a new technique based on connected components (CCs) analysis is suggested to segment the text into individual lines. The experiments were performed with two public databases: sample and test ICDAR2015 Smartphone document OCR. Experimental results demonstrate that the presented system can achieve a very good extraction rate and work efficiently even under different types of document image distortions.