S. Manjunath

Symbolic representation of text documents

This paper presents a novel method of representing a text document by the use of interval valued symbolic features. A method of classification of text documents based on the proposed representation is also presented. The newly proposed model significantly reduces the dimension of feature vectors and also the time taken to classify a given document. Further, extensive experimentations are conducted on vehicles-wikipedia datasets to evaluate the performance of the proposed model. The experimental results reveal that the obtained results are on par with the existing results for vehicles-wikipedia dataset. However, the advantage of the proposed model is that it takes relatively a less time for classification as it is based on a simple matching strategy.

Textural features in flower classification

In this work, we investigate the effect of texture features for the classification of flower images. A flower image is segmented by eliminating the background using a threshold-based method. The texture features, namely the color texture moments, gray-level co-occurrence matrix, and Gabor responses, are extracted, and combinations of these three are considered in the classification of flowers. In this work, a probabilistic neural network is used as a classifier. To corroborate the efficacy of the proposed method, an experiment was conducted on our own data set of 35 classes of flowers, each with 50 samples. The data set has different flower species with similar appearance (small inter-class variations) across different classes and varying appearance (large intra-class variations) within a class. Also, the images of flowers are of different pose, with cluttered background under various lighting conditions and climatic conditions. The experiment was conducted for various sizes of the datasets, to study the effect of classification accuracy, and the results show that the combination of multiple features vastly improves the performance, from 35% for the best single feature to 79% for the combination of all features. A qualitative comparative analysis of the proposed method with other well-known existing state of the art flower classification methods is also given in this paper to highlight the superiority of the proposed method.

Segmentation and classification of tobacco seedling diseases

In this paper, we present a novel algorithm for extracting lesion area and application of neural network to classify seedling diseases such as anthracnose and frog-eye spots on tobacco leaves. The lesion areas with anthracnose and frog-eye spots on a leaf of tobacco seedlings are segmented by contrast stretching transformation with an adjustable parameter and morphological operations. First order statistical texture features are extracted from lesion area to detect and diagnose the disease type. These texture features are then used for classification purpose. A Probabilistic Neural Network (PNN) is employed to classify anthracnose and frog-eye spots present on tobacco seedling leaves. In order to corroborate the efficacy of the proposed model we have conducted an experimentation on a dataset of 800 extracted areas of tobacco seedling leaves which are captured in an uncontrolled lighting conditions. The methodology presented herein effectively detected and classified the tobacco seedlings lesions upto an accuracy of 88.5933%. Further the recommended features are compared with Gray Level Co-occurrence Matrix (GLCM) based features to bring out their superiorities.

A non parametric shot boundary detection: an eigen gap based approach

Shot boundary detection is one of the challenging and crucial task in designing video archival and retrieval system. Even though a good number of works on shot boundary detection can be found in the literature, still there is a need of developing novel shot boundary detection algorithms which can work in a real time environment. In this paper we present a novel, simple, non parametric shot boundary detection approach where eigen gap analysis is exploited to preserve the variations among the video frames. To detect different types of shot boundaries the concept of small eigen values on a curve obtained by plotting the variations of eigen gap across the video frames is used. To corroborate the efficacy of the proposed method an experimentation on a large set of video frames having different types of transitions is carried out. For experimentation, news, lecture and entertainment videos collected from the World Wide Web is used. Also a qualitative comparative analysis of the proposed method with other existing similar type of techniques has been provided in the paper.

Feature level fusion of multi-instance finger knuckle print for person identification

The aim of this paper is to study the effect of feature level fusion of multi instances of finger knuckle prints. Initially, Zernike moments are extracted for a single instance of finger knuckle print of a person and study the identification accuracy. Subsequently, the effect of identification accuracy using feature level fusion of multi-instances of knuckle prints of a person is studied. As the length of the feature vectors of different instances of knuckle print is same, one could augment the feature vectors to generate a new feature vector. The process of concatenation of feature vectors may lead to the curse of dimensionality problem. In order to handle the curse of dimensionality, the feature dimensions are reduced prior and after the feature sets fusion using Principal Component Analysis (PCA). Experiments are conducted on PolyU finger knuckle print database to assess the actual advantage of the fusion of multi-instance knuckle prints performed at the feature extraction level, in comparison to the single instance knuckle print. Further, extensive experimentations are conducted to evaluate the performance of the proposed method against subspace methods.

Whorl identification in flower: a Gabor based approach

In this paper a novel approach for identification of whorl part of flowers useful for flower classification is presented. The problem is challenging because of the sheer variety of flower classes, intra-class variability, variation within a particular flower, and variability of imaging conditions like lighting, pose, foreshortening etc. A flower image is segmented using color information obtained using HYPEs color specifier. To identify the whorl of flowers the Gabor response of the segmented flower image is extracted and based on the Gabor response we present a method of identifying the whorl part of the flower. For experimentation we have created our own dataset of 20 classes of flowers each with 20 samples. To study the efficiency of the proposed method we have compared the obtained results with the results provided by two human experts and the results are more encouraging.

Classification of Moving Vehicles in Traffic Videos

In this paper, we propose a model for classification of moving vehicles in traffic videos. We present a corner-based tracking method to track and detect moving vehicles. The detected vehicles are classified into 4 different types of vehicle classes using optimal classifiers. The proposed classification method is based on overlapping the boundary curves of each vehicle while tracking it in sequence of frames to reconstruct a complete boundary shape of it. The reconstructed boundary shape is normalized and a set of efficient shape features are extracted. Vehicles are classified by k-NN rule and the proposed weighted k-NN classifier. Experiments are conducted on 23.02 minutes of moderate traffic videos of roadway scenes taken in an uncontrolled environment during day time. The proposed method has 94.32% classification accuracy which demonstrates the effectiveness of our method. The proposed method has 87.45% of precision with 79% recall rate for classification of moving vehicles.

Texture in Classification of Pollen Grain Images

In this paper we present a model for classification of pollen grain images based on surface texture. The surface textures of pollens are extracted using different models like Wavelet, Gabor, Local Binary Pattern (LBP), Gray Level Difference Matrix (GLDM) and Gray Level Co-Occurrence Matrix (GLCM) and combination of these features. The Nearest Neighbor (NN) classifier is adapted for classification. Unlike other existing contemporary works which are designed for a specific family or for one or few different families, the proposed model is designed independent of families of pollen grains. Experimentations on a dataset containing pollen grain images of about 50 different families totally 419 images of 18 classes have been conducted to demonstrate the performance of the proposed model. A classification rate up to 91.66 % is achieved when Gabor wavelet features are used.

Fusion of covariance matrices of PCA and FLD

In this paper, we propose a novel approach for fusing two classifiers, specifically classifiers based on subspace analysis, during feature extraction. A method of combining the covariance matrices of the Principal Component Analysis (PCA) and Fisher Linear Discriminant (FLD) is presented. Unlike other existing fusion strategies which fuse classifiers either at data level, or at feature level or at decision level, the proposed work combines two classifiers while extracting features introducing a new unexplored area for further research. The covariance matrices of PCA and FLD are combined using a product rule to preserve the natures of both covariance matrices with an expectation to have an increased performance. In order to show the effectiveness of the proposed fusion method, we have conducted a visual simulation on iris data. The proposed model has also been tested by performing clustering on standard datasets such as Zoo, Wine, and Iris. To study the versatility of the proposed method we have carried out an experimentation on sports video shot retrieval problem. The experimental results signify that the proposed fusing approach has an improved performance over individual classifiers.

Feature selection and indexing of online signatures

In this paper, we propose a model for feature selection and indexing of online signatures based person identification. For representation of online signatures, a set of 100 global features of MCYT online signature database is considered. However, MCYT based features are high dimension features which significantly increases the response time and space requirements for signature identification process. To overcome this problem, multi cluster feature selection method is proposed to reduce the dimensionality by finding a relevant feature subset. Moreover, in some applications, where the database is supposed to be very large, the identification process typically has an unacceptably long response time. A solution to speed up the identification process is to design an indexing model prior to identification which reduces the number of candidate hypotheses to be considered during matching by the identification algorithm. Hence in this paper, Kd-tree based indexing model is designed for online signatures based person identification. The experimental results reveal that the proposed model works more efficiently both in terms of time and accuracy.