Pradeep M. Patil

Offline Recognition of Devanagari Script: A Survey

In India, more than 300 million people use Devanagari script for documentation. There has been a significant improvement in the research related to the recognition of printed as well as handwritten Devanagari text in the past few years. State of the art from 1970s of machine printed and handwritten Devanagari optical character recognition (OCR) is discussed in this paper. All feature-extraction techniques as well as training, classification and matching techniques useful for the recognition are discussed in various sections of the paper. An attempt is made to address the most important results reported so far and it is also tried to highlight the beneficial directions of the research till date. Moreover, the paper also contains a comprehensive bibliography of many selected papers appeared in reputed journals and conference proceedings as an aid for the researchers working in the field of Devanagari OCR.

Rotation, scale and translation invariant handwritten Devanagari numeral character recognition using general fuzzy neural network

In this paper a general fuzzy hyperline segment neural network is proposed [P.M. Patil, Pattern classification and clustering using fuzzy neural networks, Ph.D. Thesis, SRTMU, Nanded, India, January 2003]. It combines supervised and unsupervised learning in a single algorithm so that it can be used for pure classification, pure clustering and hybrid classification/clustering. The method is applied to handwritten Devanagari numeral character recognition and also to the Fisher Iris database. High recognition rates are achieved with less training and recall time per pattern. The algorithm is rotation, scale and translation invariant. The recognition rate with ring data features is found to be 99.5%.

Robust and secured image-adaptive data hiding

Rapid growth in the demand and consumption of digital information in past decade has led to valid concerns over issues such as content security, authenticity and digital right management. Imperceptible data hiding in digital images is an excellent example of demonstration of handling these issues. Classical Cryptography is related with concealing the content of messages, whereas, Steganography is related with concealing the existence of communication by hiding the messages in cover. This paper presents a robust and secured method of embedding high volume of text information in digital Cover-images without incurring any perceptual distortion. It is robust against intentional or unintentional attacks such as image compression, tampering, resizing, filtering and Additive White Gaussian Noise (AWGN). The schemes available in the literature can deal with these attacks individually, whereas the proposed work is a single methodology that can survive all these attacks. Image Adaptive Energy Thresholding (AET) is used while selecting the embedding locations in frequency domain. Coding framework with Class Dependent Coding Scheme (CDCS) along with redundancy and interleaving of embedded information gives enhancement in data hiding capacity. Perceptual quality of images after data hiding has been tested using Peak Signal to Noise Ratio (PSNR) whereas statistical variations in selected Image Quality Measures (IQMs) are observed with respect to Steganalysis. The results have been compared with existing algorithms like STOOL in spatial domain, COX in DCT domain and CDMA in DWT domain.

Dynamic Time Warping Based Static Hand Printed Signature Verification

Static signature verication has a signicant use in establishing the authenticity of bank checks, insurance and legal documents based on the signatures they carry. As an individual signs only a few times on the forms for opening an account with any bank or for insurance related purposes, the number of genuine signature templates available in banking and insurance applications is limited, a new approach of static handwritten signature verication based on Dynamic Time Warping (DTW) by using only ve genuine signatures for training is proposed in this paper. Initially the genuine and test signatures belonging to an individual are normalized after calculating the aspect ratios of the genuine signatures. The horizontal and vertical projection features of a signature are extracted using discrete Radon transform and the two vectors are combined to form a combined projection feature vector. The feature vectors of two signatures are matched using DTW algorithm. The closed area formed by the matching path around the diagonal of the DTW-grid is computed and is multiplied with the dierence cost between the feature vectors. A threshold is calculated for each genuine sample during the training. The test signature is compared with each genuine sample and a matching score is calculated. A decision to accept or reject is made on the average of such scores. The entire experimentations were performed on a global signature database (GPDS-Signature Database) of 2106 signatures with 936 genuine signatures and 1170 skilled forgeries. To evaluate the performance, experiments were carried out with 4 to 5 genuine samples for training and with dierent ‘scores’. The proposed as well as the existing DTW-method were implemented and compared. It is observed that the proposed method is superior in terms of Equal Error Rate (EER) and Total Error Rate (TER) when 4 or 5 genuine signatures were used for training. Also it is observed that the False Acceptance Rate (FAR) of the proposed system decreases as the number of genuine training samples increases.

Automatic processing of handwritten bank cheque images: a survey

Bank cheques (checks) are still widely used all over the world for financial transactions. Huge volumes of handwritten bank cheques are processed manually every day in developing countries. In such a manual verification, user written information including date, signature, legal and courtesy amounts present on each cheque has to be visually verified. As many countries use cheque truncation systems (CTS) nowadays, much time, effort and money can be saved if this entire process of recognition, verification and data entry is done automatically using images of cheques. An attempt is made in this paper to present the state of the art in automatic processing of handwritten cheque images. It discusses the important results reported so far in preprocessing, extraction, recognition and verification of handwritten fields on bank cheques and highlights the positive directions of research till date. The paper has a comprehensive bibliography of many references as a support for researchers working in the field of automatic bank cheque processing. The paper also contains some information about the products available in the market for automatic cheque processing. To the best of our knowledge, there is no survey in the area of automatic cheque processing, and there is a need of such a survey to know the state of the art.

Database Development and Recognition of Handwritten Devanagari Legal Amount Words

A dataset containing 26,720 handwritten legal amount words written in Hindi and Marathi languages (Devanagari script) is presented in this paper along with a training-free technique to recognize such handwritten legal amounts present on Indian bank cheques. The recognition of handwritten legal amount words in Hindi and Marathi languages is a challenging because of the similar size and shape of many words in the lexicon. Moreover, many words have same suffixes or prefixes. The recognition technique proposed is a combination of two approaches. The first approach is based on gradient, structural and cavity (GSC) features along with a binary vector matching (BVM) technique. The second approach is based on vertical projection profile (VPP) feature and dynamic time warping (DTW). A number of highly matched words in both the approaches are considered for the recognition step in the combined approach based on a ranking scheme. Syntactical knowledge related to the languages is also used to achieve higher reliability. To the best of our knowledge, this is the first work of its kind in recognizing handwritten legal amounts written in Hindi and Marathi. Researchers interested in the dataset can contact the authors to get it through a shared link.

General fuzzy hyperline segment neural network

We propose a general fuzzy hyperline segment neural network (GFHLSNN) and its learning algorithm, which is an extension of the fuzzy hyperline segment neural network proposed by Kulkarni et al (2001). It combines supervised and unsupervised learning in a single algorithm so that it can be used for pure classification, pure clustering and hybrid classification/clustering.

Face Recognition by PCA Technique

Face recognition is one of the most active research areas in computer vision and pattern recognition with practical applications. This work proposes an apperence based Eigenface technique. PCA is used in extracting the relevant information in human faces. In this method the Eigen vectors of the set of training images are calculated which define the face space. Face images are projected on to the face space which encodes the variation among known face images. These encoded variations are used for recognition. Experiments are carried on IndianFace Database; the obtained recognition rate is 92.30%. The same training set is tested with nonface database.

Performance evaluation of fuzzy neural network with various aggregation operators

The modified fuzzy neural network (MFNN) proposed by Kulkarni and Sontakke is an extension of the fuzzy neural network (FNN) proposed by Kwan and Cai. Unlike FNN, the MFNN uses the Yager class of fuzzy union and intersection operators and works under a supervised environment. The paper describes MFNN with its learning algorithm. The MFNN is extended further and its performance is verified using various fuzzy aggregation operators. It is observed that Dubois and Prade operators give highest recognition rates, as compared to other operators for Fisher Iris data. Classification performance of the Hamacher operator is better with less number of neurons for Fisher Iris data, whereas the performance with the min-max operator for Wine data is better. Timing analysis is also performed and training time is found nearly equal for all the operators. The recall time per pattern is significantly less in the case of Schweizer and Sklar, and Hamacher operators. Thus, instead of using min-max or Yager class of operators one can tune the performance of the MFNN classifier to improve generalization performance by proper selection of aggregation operators.

Verification of hand printed signature images using discrete dyadic wavelet transform

An off-line hand printed signature verification method is proposed to detect skilled forgeries of amateur type by using only two genuine signature templates of an individual. Almost all other techniques in the context of off-line signature verification need at least six genuine signature templates. It is very important to note that an individual signs only twice or thrice in the application form for opening an account with a bank. The aspect ratio and discrete dyadic wavelet transforms in 2j scale of horizontal projection, vertical projection, right envelope, left envelope, top envelope and bottom envelope of the binary format of the genuine signatures are treated as the feature set of an individual. Based on these feature set the system verifies whether the signature is a forgery or not. The Dyadic wavelet used here is the quadratic spline of compact support. The matching is performed using a mixed method of correlation and distance calculation of the feature sets of genuine and skilled forgeries. Experimental results show an average verification rate of 81%.