Atul Kumar ¹, Gurpreet Singh Lehal ¹, Gurpreet Singh Lehal ¹

Affiliations
1 Department of Computer Science, Punjabi University, Patiala – 147002, Punjab, IN

Abstract

Objectives: This paper proposes a new technique for correcting errors done by Devanagari OCR (Optical Character Reader) system based on confusion matrix. Methods/Statistical Analysis: Confusion matrix is generated from large corpus of Hindi. The system takes each word of OCR output and generate number of strings from topmost five confused characters for each character of input word along with probability of these strings for ranking. Each string is validated with the character trigram dictionary and these valid strings are used for best suggestions. Findings: The topmost five words is taken as suggestions. The system has been tested for variety of OCR outputs documents of Devanagari script. The system provides suggestions for all the correct words at top position. For more than 10000 unique words in Devanagari OCR output, system gives the accuracy of 97%. Application/Improvements: This system is used in post-processing of Devanagari OCR. With some improvements, the system can also be used for Gurumukhi Script and Urdu script.

Keywords

Automatic Text Correction, Confusion Matrix, Devanagari, OCR, Trigram.

Full Text

   

Atul Kumar ¹

Affiliations
1 Department of Computer Science, Punjabi University, Patiala, Punjab, IN

Abstract

Objectives: Image Binarization is the method which converts coloured and grayscale images into black and white images, black as foreground and white as background. This paper presents the new technique for binarization of degraded documents using moving averages. Methods/Statistical Analysis: Binarization of highly degraded images is very challenging task. In the proposed technique, moving averages is calculated horizontally and vertically by setting up the value of parameters. After this, local threshold value is calculated for every pixel based on moving averages within the local rectangular block both horizontally and vertically. Findings: The proposed method has been tested with large number of degraded documents and compared with existing techniques. The proposed technique makes use of the intensity of neighbouring pixels both upward and from left to right. The proposed method produce binarization results better when compared with existing Sauvola and Niblack’s Method. Applications: The proposed method produces a bianrized image which can be used in various applications like Optical character recognition, document layout analysis.

Keywords

Binarization, Degraded Documents, Grayscale, Local Thresholding, Moving Averages.

Full Text

   

Shweta Bansal ¹, Atul Kumar ², S. S. Agrawal ¹

Affiliations
1 KIIT College of Engineering, Gurgaon - 122102, Haryana, IN
2 Ansal University, Gurgaon - 122003, Haryana, IN

Abstract

This work compares the performance of the Mel-Frequency Cepstral Coefficient (MFCC) and Perceptual Linear Prediction (PLP) features for developing a text-dependent speaker identification system. Continuously spoken Hindi speech sentences have been used to train the HMM models using HTK toolkit for each speaker separately. The experiments have been performed using a set of 200 continuously spoken sentences with vocabulary of 20000 isolated words using a database of 100 speakers. The results show an accuracy of 92.26% recognition when PLP features have been used and accuracy of 91.18% for MFCC features. A confusion matrix has been created for all the 20 test speakers based on the recognition scores obtained for each of these speakers and their confusion with other speakers. Performance has been compared in the closed set and open set conditions of testing and as it is expected, the performance in the closed set condition is far better than the open set. We propose that if PLP features are used in place of MFCC, they may provide improvement in speaker identification accuracy by reducing the cases of false acceptance.

Keywords

Hindi Speech, HMM, MFCC, RASTA-PLP, Speaker Identification.

Full Text