A. G. Ramakrishnan

ECG coding by wavelet-based linear prediction

Presents a novel coding scheme for the electrocardiogram (ECG). Following beat delineation, the periods of the beats are normalized by multirate processing. After amplitude normalization, a discrete wavelet transform is applied to each beat. Due to the period and amplitude normalization, the wavelet transform coefficients bear a high correlation across beats at identical locations. To increase the compression ratio, the residual sequence obtained after linear prediction of the significant wavelet coefficients is transmitted to the decoder. The difference between the actual period and the mean beat period, and that between the actual scale factor and the average amplitude scale factor are also transmitted for each beat. At the decoder, the inverse wavelet transform is computed from the reconstructed wavelet transform coefficients. The original amplitude and period of each beat are then recovered. The approximation achieved, at an average rate of 180 b/s, is of high quality. The authors have evaluated the normalized maximum amplitude error and its position in each cycle, in addition to the normalized root mean square error. The significant feature of the proposed technique is that, while the error is nearly uniform throughout the cycle, the diagnostically crucial QRS region is kept free of maximal reconstruction error.

Word level multi-script identification

We report an algorithm to identify the script of each word in a document image. We start with a bi-script scenario which is later extended to tri-script and then to eleven-script scenarios. A database of 20,000 words of different font styles and sizes has been collected and used for each script. Effectiveness of Gabor and discrete cosine transform (DCT) features has been independently evaluated using nearest neighbor, linear discriminant and support vector machines (SVM) classifiers. The combination of Gabor features with nearest neighbor or SVM classifier shows promising results; i.e., over 98% for bi-script and tri-script cases and above 89% for the eleven-script scenario.

Script identification in printed bilingual documents

Identification of the script of the text in multi-script documents is one of the important steps in the design of an OCR system for the analysis and recognition of the page. Much work has already been reported in this area relating to Roman, Arabic, Chinese, Korean and Japanese scripts. In the Indian context, though some results have been reported, the task is still at its infancy. In the work presented in this paper, a successful attempt has been made to identify the script, at the word level, in a bilingual document containing Roman and Tamil scripts. Two different approaches have been proposed and thoroughly tested. In the first method, words are divided into three distinct spatial zones. The spatial spread of a word in upper and lower zones, together with the character density, is used to identify the script. The second technique analyses the directional energy distribution of a word using Gabor filters with suitable frequencies and orientations. Words with various font styles and sizes have been used for the testing of the proposed algorithms and the results are quite encouraging.

Automation of differential blood count

A technique for automating the differential count of blood is presented. The proposed system takes, as input, color images of stained peripheral blood smears and identifies the class of each of the white blood cells (WBC), in order to determine the count of cells in each class. The process involves segmentation, feature extraction and classification. WBC segmentation is a two-step process carried out on the HSV-equivalent of the image, using k-means clustering followed by the EM-algorithm. Features extracted from the segmented cytoplasm and nucleus, are motivated by the visual cues of shape, color and texture. Various classifiers have been explored on different combinations of feature sets. The results presented are based on trials conducted with normal cells. For training the classifiers, a library set of 50 patterns, with about 10 samples from each class, is used. The test data, disjoint from the training set, consists of 34 patterns, fairly represented by every class. The best classification accuracy of 97% is obtained using neural networks, followed by 94% using SVM.

Machine recognition of online handwritten Devanagari characters

In this paper, we describe a system for the automatic recognition of isolated handwritten Devanagari characters obtained by linearizing consonant conjuncts. Owing to the large number of characters and resulting demands on data acquisition, we use structural recognition techniques to reduce some characters to others. The residual characters are then classified using the subspace method. Finally the results of structural recognition and feature-based matching are mapped to give final output. The proposed system is evaluated for the writer dependent scenario.

Comparison of elastic matching algorithms for online Tamil handwritten character recognition

We present a comparison of elastic matching schemes for writer dependent on-line handwriting recognition of isolated Tamil characters. Three different features are considered namely, preprocessed x-y coordinates, quantized slope values, and dominant point coordinates. Seven schemes based on these three features and dynamic time warping distance measure are compared with respect to recognition accuracy, recognition speed, and number of training templates. Along with these results, possible grouping strategies and error analysis is also presented in brief.

Epoch Extraction Based on Integrated Linear Prediction Residual Using Plosion Index

Epoch is defined as the instant of significant excitation within a pitch period of voiced speech. Epoch extraction continues to attract the interest of researchers because of its significance in speech analysis. Existing high performance epoch extraction algorithms require either dynamic programming techniques or a priori information of the average pitch period. An algorithm without such requirements is proposed based on integrated linear prediction residual (ILPR) which resembles the voice source signal. Half wave rectified and negated ILPR (or Hilbert transform of ILPR) is used as the pre-processed signal. A new non-linear temporal measure named the plosion index (PI) has been proposed for detecting ‘transients’ in speech signal. An extension of PI, called the dynamic plosion index (DPI) is applied on pre-processed signal to estimate the epochs. The proposed DPI algorithm is validated using six large databases which provide simultaneous EGG recordings. Creaky and singing voice samples are also analyzed. The algorithm has been tested for its robustness in the presence of additive white and babble noise and on simulated telephone quality speech. The performance of the DPI algorithm is found to be comparable or better than five state-of-the-art techniques for the experiments considered.

Eye detection using color cues and projection functions

We propose a heuristic approach for detection of eyes in close-up images. The experimental images are stereotypical mug shot faces which can be expected in applications of face recognition systems, say in ATM vestibules. We prove the efficacy of our proposed method in detection of eyes, both in indoor and outdoor environments with considerable scale variations and an allowable rotation in the image plane. We employ a hierarchical search space reduction technique to localize possible eye areas in the image. The distinct human skin color and low intensity areas of the eye ball are the primary cues used to locate eye regions. Further on, eye validation is performed using the mean and variance projection functions.

Gabor filters for document analysis in Indian bilingual documents

Reasonable success has been achieved at developing monolingual OCR systems in Indian scripts. Scientists, optimistically, have started to look beyond. Development of bilingual OCR systems and OCR systems with capability to identify the text areas are some of the pointers to future activities in Indian scenario. The separation of text and non-text regions before considering the document image for OCR is an important task. In this paper, we present a biologically inspired, multi-channel filtering scheme for page layout analysis. The same scheme has been used for script recognition as well. Parameter tuning is mostly done heuristically. It has also been seen to be computationally viable for commercial OCR system development.

Fetal lung maturity analysis using ultrasound image features

This pilot study was carried out to find the feasibility of analyzing the maturity of the fetal lung using ultrasound images. Data were collected from normal pregnant women at intervals of two weeks from the gestation age of 24 to 38 weeks. Images were acquired at two centers located at different geographical locations. The total data acquired consisted of 750 images of immature and 250 images of mature class. A region of interest of 64/spl times/64 pixels was used for extracting the features. Various textural features were computed from the fetal lung and liver images. The ratios of fetal lung to liver feature values were investigated as possible indexes for classifying the images into those from mature (reduced pulmonary risk) and immature (possible pulmonary risk) lung. The features used are fractal dimension, lacunarity, and features derived from the histogram of the images. The following classifiers were used to classify the fetal lung images as belonging to mature or immature lung: nearest neighbor, k-nearest neighbor, modified k-nearest neighbor, multilayer perceptron, radial basis function network, and support vector machines. The classification accuracy obtained for the testing set ranges from 73% to 96%.