Atul Negi

An OCR system for Telugu

Telugu is the language spoken by more than 100 million people of South India. Telugu has a complex orthography with a large number of distinct character shapes (estimated to be of the order of 10,000) composed of simple and compound characters formed from 16 vowels (called achchus) and 36 consonants (called hallus). We present an efficient and practical approach to Telugu OCR which limits the number of templates to be recognized to just 370, avoiding issues of classifier design for thousands of shapes or very complex glyph segmentation. A compositional approach using connected components and fringe distance template matching was tested to give a raw OCR accuracy of about 92%. Several experiments across varying fonts and resolutions showed the approach to be satisfactory.

Zone identification in the printed Gujarati text

Gujarati, is a language from the Indo-Aryan family of languages, used by 50 million people in the western part of India. Gujarati-script used to write the Gujarati language, is a multilevel script, written in three zones: base character zone, upper modifier zone and lower modifier zone. Several characters are discriminated by the specific modifiers, which exist in the upper and lower zones. Hence, detecting the zone boundaries is an important task in the Gujarati OCR. Although the Gujarati script is in some respects related to the Devanagari script, there are certain peculiar differences, which prevent the use of already known techniques for zone boundary detection for scripts such as Bengali, Assamese and Devanagari where mature OCR systems already do exist. There is only one previous documented effort for Gujarati OCR, in which an approach to recognize a small subset of Gujarati alphabet was discussed. The present paper proposes a sophisticated method for accurate zone detection in images of printed Gujarati. It is expected that this approach shall make the way smoother for the design and development of Gujarati OCR systems for complete character sets.

SubXPCA and a generalized feature partitioning approach to principal component analysis

In this paper we propose a general feature partitioning framework to PCA computation and raise issues of cross-sub-pattern correlation, feature ordering dependence, selection of sub-pattern size, overlap of sub-patterns and selection of principal components. These issues are critical to the design and performance of feature partitioning approaches to PCA computation. We show several open issues and present a novel algorithm, SubXPCA which proposes a solution to the cross-sub-pattern correlation issue in the feature partitioning framework. SubXPCA is shown to be a general technique since we derive PCA and SubPCA as special cases of SubXPCA. We show SubXPCA has theoretically better time complexity as compared to PCA. Comprehensive experimentation on UCI repository data and face data sets (ORL, CMU, Yale) confirms the superiority of SubXPCA with better classification accuracy. SubXPCA not only has better time performance but is also superior in its summarization of variance as compared to SubPCA. SubXPCA is shown to be robust in its performance with respect to feature ordering and overlapped sub-patterns.

Localization, extraction and recognition of text in Telugu document images

In this paper we present a system to locate, extract andrecognize Telugu text. The circular nature of Telugu scriptis exploited for segmenting text regions using the HoughTransform. First, the Hough Transform for circles is performedon the Sobel gradient magnitude of the image tolocate text. The located circles are filled to yield text regions,followed by Recursive XY Cuts to segment the regionsinto paragraphs, lines and word regions. A regionmerging process with a bottom-up approach envelopes individualwords. Local binarization of the word MBRs yieldsconnected components containing glyphs for recognition.The recognition process first identifies candidate charactersby a zoning technique and then constructs structural featurevectors by cavity analysis. Finally, if required, crossingcount based non-linear normalization and scaling is performedbefore template matching. The segmentation processsucceeds in extracting text from images with complexNon-Manhattan layouts. The recognition process gave acharacter recognition accuracy of 97%-98%.

Principle application and vision in Internet of Things (IoT)

Internet of Things (IoT) is fast becoming a disruptive technology business opportunity, with standards emerging primarily for wireless communication between sensors, actuators and gadgets in day-to-day human life, all in general being referred to as “Things”. This offers the capability to measure for understanding environment indicators. This paper addresses the internet of things (IoT) as the main enabling factor of promising paradigm for integration and comprehensive of several technologies for communication solution, Identification and integrating for tracking of technologies as wireless sensor and actuators. IoT as envisioned is billion sensors connected to the internet through the sensors that would be generate large amount of data which need to analyzed, interpreted and utilized. Context aware capturing enables modeling, interpreting and storing of sensor data which is linked to appropriate context variable dynamically. Building or home automation, social smart communication for enhancement of quality of life, that could be considered as one of the application of IoT where the sensors, actuators and controllers can be connected to internet and controlled. This paper introduces the concept of application for internet of things and with the discussion of social and governance issues that arise as the future vision of internet of things.

Mutual Authentication and Key Agreement based on Elliptic Curve Cryptography for GSM

In this paper we propose an improvement to the GSM authentication protocol, based on Elliptic Curve Cryptography. The proposed protocol offers enhanced security since it does not use A5/0, A5/1 and A5/2 algorithms which have been already broken. The proposed protocol provides mutual authentication, requires less storage, avoids replay attack and consumes smaller network bandwidth.

A survey of distance/similarity measures for categorical data

Similarity or distance between two objects plays a fundamental role in many data mining tasks like classification and clustering. Categorical data, unlike numeric data, conceptually is deficient of default ordering relations on the attribute values. This makes the task of devising similarity or distance metrics and data mining tasks such as classification and clustering of categorical data more challenging. In this paper we formulate a taxonomy of various distance or similarity measures used in conjunction with data whose attributes are categorical. We categorize the existing measures into two broad classes, namely, Context-free and Context-sensitive measures for categorical data. In addition, we suggest a taxonomy of the clustering approaches for categorical data. We also propose a hybrid approach for measuring similarity between objects. We make a relative comparison of the strengths and weaknesses of some of the similarity measures and point out future research directions.

Smart Replica Selection for Data Grids Using Rough Set Approximations (RSDG)

The best replica selection problem is one of the important aspects of data management strategy of data grid infrastructure. Recently, rough set theory has emerged as a powerful tool for problems that require making optimal choice amongst a large enumerated set of options. In this paper, we propose a new replica selection strategy using a grey-based rough set approach. Here first the rough set theory is used to nominate a number of replicas, (alternatives of ideal replicas) by lower approximation of rough set theory. Next, linguistic variables are used to represent the attributes values of the resources (files) in rough set decision table to get a precise selection cause, some attribute values like security and availability need to be decided by linguistic variables (grey numbers) since the replica mangers’ judgments on attribute often cannot be estimated by the exact numerical values (integer values). The best replica site is decided by grey relational analysis based on a grey number. Our results show an improved performance, compared to the previous work in this area.

Wavelet Feature Based Confusion Character Sets for Gujarati Script

Indic script recognition is a difficult task due to the large number of symbols that result from concatenation of vowel modifiers to basic consonants and the conjunction of consonants with modifiers etc. Recognition of Gujarati script is a less studied area and no attempt is made so far to constitute confusion sets of Gujarati glyphs. In this paper, we present confusion sets of glyphs in printed Gujarati. Feature vector made up of Daubechies D4 wavelet coefficients were subjected to two different classifiers, giving more than 96% accuracy for a larger set of symbols. Novel application of GR neural-net architecture allows for fast building of a classifier for the large character data set. The combined approach of wavelet feature extraction and GRNN classification has given the highest recognition accuracy reported on this script.

On developing high accuracy OCR systems for Telugu and other Indian scripts

In this paper we list a number of factors that are important in achieving high recognition accuracy in OCR systems for Telugu and other Indian scripts. While it is relatively easy to obtain 85%-93% accuracy, it becomes increasingly difficult to improve the performance further We discuss how the factors presented in this paper helped achieve an accuracy of nearly 97% with our OCR system for Telugu script. It is expected that these factors are specific not only to Telugu but also work for other Indian scripts in general and south Indian scripts in particular.