Optical character recognition based on local invariant features

ABSTRACT

The Optical Character Reader (OCR) process means the transition from scanned manual or written images to a machine-determined document. The American Standard Code for Information Interchange (ASCII) in cognitive processing uses OCR. The challenge is two primary folds: word segmentation by letters and character recognition. Implement a new approach to include the two functions by Scale-Invariant Transforming Feature (SIFT) descriptors. To compare SIFT descriptors (RootSIFT), devise a new procedure, that offers outstanding results without increasing computation or storage requirements. In order to identify English characters, Artificial Bee Colony (ABC) method suggests that the back propagation neural network for classification of character be utilized. Conducted experiments with more than 10 measures intended for every character and tested the accuracy for numerical numbers, chart letters, small letters and alphanumeric characters. The performance analysis of ABC optimized neural network algorithm has achieved a maximum accuracy of 97.3077% compared with precision recall and f-measure.

KEYWORDS:

• Optical character reader (OCR)
• American Standard Code for Information Interchange (ASCII)
• scale-invariant transforming feature (SIFT)
• RootSIFT
• Artificial Bee Colony (ABC)
• back propagation neural network (BPNN)
• neural network (NN)
• positive predictive values (PPV)

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes on contributors

Sandhya Balakrishnan Poodikkalam received BSc Degree in Chemistry from Unity Women’s College, Manjeri affiliated to Calicut University in 2000, MSc Degree in Computer Science from M.E.S Kalladi College, Mannarkkad affiliated to Calicut University in 2002, and BEd in Computer Science from Vatakara BEd Center affiliated to Calicut University in 2005. She is currently working towards the PhD degree at the Department of Computer Science, CMS College of Science & Commerce, Chinnaveedampatti affiliated to Bharathiyar University. Her Research interests include OCR applications in Digital Image Processing.

Dr Loganathan Pavithira currently working as associate professor, Department of Computer Science in Dr. NGP Arts and Science College, Coimbatore has nearly 15 years of teaching experience in Computer Science. She obtained her BSc degree (Computer Science) in 2001, Master of Computer Applications in 2004 from Bharathiar University. She obtained her master of Philosophy in Computer Science from Bharathiar University during 2006 and PhD in CS from Bharathiyar University in 2015. She has published 15 papers in various international journals and presented 10 papers in various international conferences.