In this work, we present a fovea center localization method in digital color photographs of the retina (color eye fundus images). The proposed method is based on prior knowledge of optic disk center and optic disk diameter. The detection of this anatomical feature is a prerequisite for the computer aided diagnosis of several retinal diseases, such as Diabetic Maculopathy. The proposed method is evaluated on 33 retinal photographs from public DRIVE data set. The experimental results demonstrate that the proposed method is able to detect the fovea center by providing encouraging results.

Accurate and reliable segmentation is an essential step in determining valuable quantitative information on size, shape and texture, which may assist microbiologists in their diagnoses. The snakes or active contours are used extensively in computer vision and image processing applications, particularly to locate object boundaries. The objective of the present study is to develop an automatic tool to identify and classify the virus particles in digital microscopic images using multigrid active contour model. Geometric features are used to identify the different types of virus particles, namely, Rotavirus and Adenovirus using 3classifier, K-NN classifier and Neural Network classifiers. The current methods rely on the subjective reading of profiles by a human expert based on the various manual staining methods. In this paper, we propose a method for virus particle classification by segmenting digital microscopic virus images and extracting geometric features for virus particle classification. The experimental results are compared with the manual results obtained by the microbiology expert and demonstrate the efficacy of the proposed method.

Face detection is a necessary first-step in face recognition systems, with the purpose of localizing and extracting the face region from the background. It also has several applications in areas such as content-based image retrieval, video coding, video conferencing, crowd surveillance, and intelligent human–computer interfaces. In this paper, we propose a novel approach for the detection of human face in a digital image based on the fuzzy spatial interrelationships of only the prominent facial features of the face, namely, eyes and mouth. A fuzzy face model is constructed for the face detection algorithm. The experimentation has been done using several face databases. The experimental results show that the proposed algorithm performs satisfactorily with an average accuracy of 96.10% and is efficient in terms of accuracy and detection time despite the exclusion of other facial features, namely, nose, eyebrows and ears.

In this paper, we present an effective system for recognizing flower images taken by digital cameras. A flower image is segmented by eliminating the background using a Iterated Graph Cut method. We obtained the skeleton from the segmented flower images using skeleton pruning method. The shape context feature are extracted from skeleton of flower images. In this work, nearest neighbor is used as a classifier. To corroborate the efficacy of the proposed method, an experiment was conducted on our own data set of 30 classes of flowers, containing 3000 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. In addition, the images of flowers are of different poses, with cluttered background under different lighting and climatic conditions. An experiment has been conducted by picking images randomly from the database, it is shown that relatively a good performance can be achieved, using shape context features with the nearest neighbor classifier algorithm.

Image segmentation has long been a basic problem in computer vision. Depth-wise Layering is a kind of segmentation that slices an image in a depth-wise sequence unlike the conventional image segmentation problems dealing with surface-wise decomposition. The proposed Depth-wise Layering technique uses a single depth image of a static scene to slice it into multiple layers. The technique employs a thresholding approach to segment rows of the dense depth map into smaller partitions called Line-Segments in this paper. Then, it uses the line-segment labelling method to identify number of objects and layers of the scene independently. The final stage is to link objects of the scene to their respective object-layers. We evaluate the efficiency of the proposed technique by applying that on many images along with their dense depth maps. The experiments have shown promising results of layering.

In this paper, a new algorithm for segmentation of lesions on tobacco seedling leaves is proposed. Segmentation algorithm consists of mainly two steps. First step is to approximate lesion extraction using contrast stretching transformation and morphological operations such as erosion and dilation. Second step refines the outcome of first step by color segmentation using CIELAB color model. We have conducted a performance evaluation of segmentation algorithm by measuring the parameters such as Measure of overlapping (MOL), Measure of under-segmentation (MUS), Measure of over-segmentation (MOS), Dice similarity measure (DSM), Error-rate (ER), Precision (P) and Recall (R). Then 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 1000 extracted areas of tobacco seedling leaves which are captured in an uncontrolled lighting conditions. Experimental results show that the proposed segmentation algorithm achieved best average DSM and MOL accuracy. The methodology presented herein effectively detected and classified the tobacco seedlings lesions upto an accuracy of 91.4412%. Further the recommended features are compared with Gray Level Co-occurrence Matrix (GLCM) based features to bring out their superiorities.

Medical image segmentation is a processing step for image understanding and computer aided diagnosis. In this paper, we propose medical image texture segmentation using Gabor filter. Image enhancement techniques are utilized to remove strong speckle noise as well to enhance the weak boundaries of medical images. We propose to exploit the concept of Gabor filtering to extract the texture content of medical images. Experiment is conducted on ImageCLEF2010 database. Results show the efficacy of our proposed medical image segmentation.

The underwater images usually suffers from non-uniform lighting, low contrast, blur and diminished colors. In this paper, we proposed an image based preprocessing technique to enhance the quality of the underwater images. The proposed technique comprises a combination of four filters such as homomorphic filtering, wavelet denoising, bilateral filter and contrast equalization. These filters are applied sequentially on degraded underwater images. The literature survey reveals that image based preprocessing algorithms uses standard filter techniques with various combinations. For smoothing the image, the image based preprocessing algorithms uses the anisotropic filter. The main drawback of the anisotropic filter is that iterative in nature and computation time is high compared to bilateral filter. In the proposed technique, in addition to other three filters, we employ a bilateral filter for smoothing the image. The experimentation is carried out in two stages. In the first stage, we have conducted various experiments on captured images and estimated optimal parameters for bilateral filter. Similarly, optimal filter bank and optimal wavelet shrinkage function are estimated for wavelet denoising. In the second stage, we conducted the experiments using estimated optimal parameters, optimal filter bank and optimal wavelet shrinkage function for evaluating the proposed technique. We evaluated the technique using quantitative based criteria such as a gradient magnitude histogram and Peak Signal to Noise Ratio (PSNR). Further, the results are qualitatively evaluated based on edge detection results. The proposed technique enhances the quality of the underwater images and can be employed prior to apply computer vision techniques.

In this paper high-level image classes are inferred from low-level image features like color and shape features with the help of artificial neural network. Back propagation neural network algorithm is used for integrating knowledge from low-level image features and classify the images into high level concepts / semantic classes. The classifier is evaluated on a database of 1000 images from COREL database. The experimental results show that the accuracy using back propagation neural network algorithm to classify COREL images ranges between 80.5% to 88.6%.

Incorporation of video facility in mobile devices has opened up various challenging research problems in the field of image processing. One such challenge is to mosaic video containing the document source. In this paper we propose an approach to mosaic the fragmented video frames containing text for the purpose of text localization and recognition by decomposing the fragmented frames into vertical strips and then matching the strips of consecutive frames to stitch the content. Similarity between strips are found using Scale Invariant Feature Transform - SIFT match algorithm as it is invariant to scale, rotation and geometric distortions like blurring / resampling of local image orientation planes. The process of blending the content of the matched strips is done with the help of a translation function on the best matched points estimated by homography using RANSAC were wrapped using translation function.

In this paper we propose two different approaches to segment and extract moving vehicles in traffic videos. Background subtraction is used to extract foreground frames. Different types of moving vehicles are then segmented by first proposed hybrid approach which combines a connected component analysis and a semi-supervised thresholding. In the second proposed approach, Gabor filtering method is used for segmentation of moving vehicles. The robustness and efficacy of our proposed approaches are elaborated by experiments conducted on real traffic videos captured under complex background, variations in illumination, motion, position of a camera and different moving directions during day time. The presented results are as well compared with 2 well-known methods of GMM and W4 used in extraction of moving vehicles.

In this paper, we address the problem of shot boundary detection, which is an essential pre-processing step in video analysis applications. We present a novel model for detection of shot boundaries based on eigen conjugation of adjacent frames. Also, in this paper we present a method of classifying the shot boundaries using template matching technique. Experimentation is carried out on two different categories of videos such as entertainment videos and sports videos.

Text block verification is important in enhancing the text detection accuracy in natural scene images because it is hard to develop general or objective heuristics to differentiate text and non-text block. In this paper, we propose new objective heuristics to verify the blocks detected by the text detection method based on angle information and proximity of the blocks. The angle for the detected block is computed using PCA to find the direction of the text block. In the same way, the proximity between pixels in the detected block is estimated to find closeness between pixels. Then the method combines these two heuristics to verify the text block to obtain a better result. We conduct experimental results on different databases to show that the performance of the text detection method increases in terms of recall, precision and f-measure with the text block verification methods. The database includes benchmark database ICDAR-2003 competition data, our own data captured by high resolution camera and captured by low resolution mobile camera.

A system for segmentation of online handwritten Bangla (or Bengali) cursive words into individual character in unconstrained domain is described in this paper. For segmentation of touching characters, we discovered some rules analyzing possible joining patterns of characters and modifiers of Bangla. We selected a set of 2000 Bangla words written by different groups of people such that they contain all basic characters, all vowel and consonant modifiers and almost all types of possible joining among them. We achieved correct segmentation rate of 97.67% on a dataset of 2000 words.

In this paper, a recognition system for handwritten and printed Kannada numerals and vowels is proposed based on zone features. The Kannada numerals and vowels are circular in nature; therefore the pixel density feature is potential features for handwritten and printed Kannada numerals and vowels. The preprocessed image is divided into 64 zones. The pixel density is computed between each zone. The dimension of feature vector is 64. For classification, SVM classifier with five fold cross validation test is used to obtain average percentage of recognition accuracy of 97.40% and 95.90% for mixture of handwritten and printed Kannada numerals and vowels respectively. The total number of classes is 48, but in this experiment 24 classes are reduced due to mixture of handwritten and printed Kannada numerals and vowels. The novelty of the proposed algorithm is free from thinning and slant of the numerals and vowels.

Vision plays an important role in perception to enable communication in a better way either in Human-Human or Human-Robot interaction system. Visual attention enhances the understanding of intent in communication, e.g. Eye gaze, orientation of face, etc. In this paper, we propose an intelligent vision system that tracks the human face. To realize the system we integrate Viola Jones Face detector, Eye detector and the Camshift algorithm. Camshift algorithm relies on back projected probabilities and it can fail to track the object due to appearance change caused by background, camera movement and illumination. Eye detector is used as verifier while initializing the Camshift algorithm and later in face tracking. The proposed system is implemented on Lego Mindstorm NXT® Robot platform and good tracking results were obtained, in the sense that the Robot and the camera were able to position in such a way that the frontal face is contained.

This paper discusses the implementation details of computer vision based gesture processing. Finger gestures are processed and recognized for interacting with the desktop and desktop applications. Gestures are captured via webcam; and conventional image processing techniques and algorithms are used for video frame analysis. For identifying the object of interest in each video frame BLOB colour-segmentation algorithm is used. 8-directional code features are extracted from the motion vector for recognizing of gestures.

This paper discusses the implementation details of computer vision based gesture processing. Finger gestures are processed and recognized for interacting with the desktop and desktop applications. Gestures are captured via webcam; and conventional image processing techniques and algorithms are used for video frame analysis. For identifying the object of interest in each video frame BLOB colour-segmentation algorithm is used. 8-directional code features are extracted from the motion vector for recognizing of gestures.

Face is the most visible trait of a person. With increasing use of biometrics, there is a real threat for the conventional systems using face databases, which store images of users in raw and unaltered form. If compromised not only it is irrevocable, but can also be misused for cross-matching across different databases. So it is desirable to generate revocable templates for the same user in different applications to prevent cross-matching and to enhance security, while maintaining privacy and ethics. In this paper we propose a novel method to address those issues.

This paper is addressing the combination of physiological and behavioral biometrics. In this we adopted multimodal approach on combining signature and fingerprint biometrics using feature level fusion with different normalization techniques to improve the performance of multimodal system. Texture features are extracted for fingerprint, pen pressure, azimuth, and altitude features are extracted for signature. Thus, classification is done using K-NN and SVM classifiers; from experimental results of two classifiers shows that multimodal approach performs well by adopting robust feature normalization technique with suitable distance measures and kernels of classifiers.

Biometric technologies are becoming the basis for highly secure identification and personal verification solutions.Hand geometry biometric identification system uses the geometric shape of the hand to identify the person. A novel graph theoretic approach to hand geometry biometrics is presented in this paper. The user hand is represented as weighted undirected complete connected graph and spectral properties of the graph are used as features vectors. User identification is performed using multilayer feed-forward neural network. Experiments show that the proposed method when tested on GPDS150 peg-free hand database of users’ right hand offers a promising result of 97.05% correct personal identification.

In this paper, a model for representation and indexing of online signatures for person identification is proposed. 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. In this paper, we study the suitability of Kd-tree indexing mechanism for person identification based on online signatures. For representation of online signatures, we considered a set of 100 global features of (MCYT online signature database) online signature and index by Kd-tree. Experimental results reveal that indexing prior to identification is faster than conventional identification method in terms of time for online signatures.

In this paper, we present a method of representing a gait by interval-valued symbolic feature for gait recognition. The proposed symbolic representation is capable of capturing variations in gait features due to change in cloth, carrying bag and different normal conditions (change in time) more effectively. A method of selecting a common threshold for validating the belongingness of a test gait to a specific class of gait is investigated. An appropriate method of matching a test gait sequence with a reference gait sequence for the purpose of gait recognition has been studied. An experimentation is conducted on a standard considerably large gait database (CASIA DATASET B) to study the efficacy of the proposed symbolic representation. The results obtained are satisfactory and encouraging.

In this paper, we propose a PCA+LDA on wavelet co-occurrence histogram features (WCHF) for texture classification. The texture features are extracted based on the co-occurrence histograms of wavelet decomposed images. The features extracted by this method form a feature vector of high dimensionality of 384 for the gray scale image. A combination of Principal Component Analysis (PCA) and Linear Discriminate Analysis (LDA) was applied on feature vector for dimensionality reduction and to enhance the class separability respectively. By applying PCA to the feature vectors, low dimensionality feature sets were obtained and processed using LDA. The vectors obtained from the LDA are representative of each texture. The classification performance is tested on a set of 32 Brodatz textures. The results are compared with the method proposed by P.S.Hiremath and S.Shivashankar [16]. It is evident from the experimental results that the proposed method exhibits superior performance in the reduced feature set.

Defects detection is very important to guarantee the welding quality. Many researchers have been done on this area. However, most of them are limited in efficiency. Different from these researches, a method that is used to detect defects from the radiographic images of welding is put forward. This paper discusses an effective method to segment of the defects in welding images. We also use Two Dimensional Left Median Filter for enhancing the images. Experimental results show this method is effective in noisy and low contrasted radiographic images of weld.

Wavelet transformation is one of the most effective mathematical tools for analyzing mammogram images which posses’ fuzzy likes texture characteristics. In this paper we carried out a comparative study of performance of discrete wavelet transformation (DWT) and stationary wavelet transformation (SWT) for classifying mammogram images into Normal, Benign and Malignant. In each wavelet transformations, a fractional part of the highest wavelet coefficients is used as features for classification. Initially we created a class core vector for each risk level using ten percent of images from each set. This acts as the basis of the classification. Then each test image in the dataset is classified into the appropriate risk level by the Euclidean distance between the features of the test image and the class core vectors. Using discrete wavelet transformation, 83 % of the images were correctly classified into exact risk level. On the other hand using stationary wavelet transformation obtained only 76% of accuracy. We also made a comparative analysis of other distance measure called Bray Curtis. But the result obtained in Bray Curtis is not much promising. The study also reveals that the redundant nature of coefficients in stationary wavelet transformation is not suitable for identifying tumors in mammograms.

This paper introduces an efficient method for recognizing spoken digits using a combination of Wavelet Packet Decomposition (WPD) and Artificial Neural Networks (ANN) classifier. Speech recognition is a fascinating application of Digital Signal Processing. There has been lot of research in the area of speech recognition for different languages. Digits in Malayalam, which belong to one of the four Dravidian languages of Southern India, are used to create the database. Wavelet Packet Decomposition is used for feature extraction in the time-frequency domain. Training, testing and pattern recognition are performed using Artificial Neural Networks (ANN). Due to the multi-resolution characteristics and efficient time frequency localizations, wavelets are very much suitable for processing non stationary signals like speech. ANNs are utilized in this work due to their parallel distributed processing, distributed memories, error stability, and pattern learning distinguishing ability. The experimental results show the effectiveness of this hybrid architecture in recognizing speech.

An electrocardiogram (ECG) is a bioelectrical signal which records the heart’s electrical activity versus time. The interpretation of ECG signal is an application of pattern recognition. The techniques used in this paper comprise: signal pre-processing, R peak detection, QRS reconstruction, RR interval detection, feature extraction and linear classifier model versus ensemble of classifier model. The processed signal source came from the Massachusetts Institute of Technology Beth Israel Hospital (MIT-BIH) arrhythmia database which was developed for research in cardiac electro-physiology. The results of recognition rates are compared to find a better structure for ECG classification. Among different classifier model, it was found that ensemble of classifier with DECORATE meta-learner model possessed the best performance with highest recognition rate of 90.36% for cardiac conditions and moderate level of agreement between computerized prediction and cardiologist interpretation. Based on this result, the method of using important ECG features plus a suitable ensemble of classifier model outperforms the single classifier model and which can increase the testing speed and the accuracy rate.

Studies have shown that human ear is one of the representative human biometrics with uniqueness and stability. Automatic Localization of 2-D ear from a side face image is one of the challenging problems. This paper presents an efficient technique for automatic ear localization from a side face image. The localization is done using an effective Skin Segmentation algorithm and Template Matching using Correlation Coefficients. The ear is localized automatically through Skin Segmentation followed by automatic ear localization. The proposed technique is tested using an ear database which contains 100 ear images and 95% of the images responded with correct automatic ear localization.

Detection and classification of spiculated masses in ultrasound images is still a challenge due to the interference of speckle noise and fuzziness of boundaries. Ultrasound (US) is an important adjunct to mammography in breast cancer detection as it doubles the rate of detection in dense breasts do a dynamic analysis of moving structures in breast. This paper presents technique to detect spiculations and boundary of spiculated masses in breast ultrasound images. In the proposed method, ultrasound images are preprocessed using Gaussian smoothing to remove additive noise and anisotropic diffusion filters to remove multiplicative noise (speckle noise). Active contour method has been used to extract a closed contour of filtered image which is the boundary of the spiculated mass. Spiculations which make breast mass unstructured or irregular are marked by measuring the angle of curvature of each pixel at the boundary of mass. To classify the breast mass as malignant or benign we have used the structure of mass in accordance with spiculations and elliptical shape. We have used receiver operating characteristic curve (ROC) to evaluate the performance. We have validated the proposed algorithm on 100 sub images(40 spiculated and 60 non spiculated) and results shows 90.5% of sensitivity with 0.87 Area Under Curve. Proposed techniques were compared and contrasted with the existing methods and result demonstrates that proposed algorithm has successfully detected spiculated mass ROI candidates in breast ultrasound images.

Transliteration is mapping of pronunciation and articulation of words written in one script into another script. Transliteration should not be confused with translation, which involves a change in language while preserving meaning. CLIR is the acronym of a great variety of techniques, systems and technologies that associate information retrieval (normally from texts) in multilingual environments. Dictionaries have often been used for query translation in cross language information retrieval (CLIR). However, we are faced with the problem of translating Names and Technical Terms from English to Kannada/Telugu. The most important query words in information retrieval are often proper names. We present a method for automatically learning a transliteration model from a sample of name pairs in two languages.

Web access pattern mining is an application of sequence mining on web log data to generate interesting user access behavior on World Wide Web. In this paper we present a new method for the efficient mining of maximal web access patterns. The method is a variation of recently published, FOL-Mine (First Occurrence List Mine) [1] for mining web access patterns. It is a top-down method that uses the concept of first occurrence to reduce search space and thus improving the performance.

In this paper we propose a novel approach of feature ranking for feature selection. This method is particularly useful for applications handling high dimensional datasets such as machine learning, pattern recognition and signal processing. This process is also applicable to small and medium sized datasets to identify significant features or attributes for a particular domain using the information contained in the dataset alone and hence the method preserves the meaning of the existing features. With the help of the proposed method, redundant attributes can be removed efficiently without sacrificing the classification performance. In this approach, after eliminating the outlier data elements from the dataset, features are ranked to identify the predominant features of the dataset. The discernibility matrix in RST is used as a tool to discover the data dependencies existing between various features and features are ranked based on these data dependencies. A method using Centre of Gravity (CoG) line is suggested to determine this discrimination frequency within a reduced computational effort. To evaluate the performance of the algorithm, we applied the proposed algorithm on a test dataset consisting of 3000 offline handwritten samples of 10 Tamil characters. The outcome of the experiment shows that the new method is efficient and effective for dimensionality reduction.

Delivering Web pages to Small Screen Devices such as Mobile Devices, Personal Digital Assistants (PDA) etc., has become possible with the latest wireless technology. However, these devices have very small screen sizes, memory capacities and low bandwidth. Today most of the Web pages are designed for Large Screen Devices, which makes browsing on Small Screen Devices extremely difficult. Therefore, a method to reconstruct Large Screen Devices optimized Web pages for Small Screen Devices is essential. Proposed methods which involves segment the Web page based on its structure, followed by noise removal based on block features and to utilize the hierarchy of the content element to regenerate a page suitable for Small Screen Devices. In this article we give a brief overview of existing approaches, their advantages and challenges. Finally we give an overview of comparison of results.

Data authentication and data security are the primary requirement in present day communication system. In image processing, data authentication is implemented using watermarking technique. Recently several watermarking techniques have been proposed. The drawback of these techniques is less robustness, less fidelity and degradation in image quality. Watermarking applications are used in copyright protection, broadcast monitoring, usage control, and this work is implemented for copyright protection using singular value decomposition (SVD) technique. In this technique, the singular values of watermark image are embedded into singular values of host image to get watermarked image. Singular value decomposition (SVD based watermarking techniques is implemented for gray scale and color image of different sizes and the comparative analysis is made based on the results produced when the image is subjected to different types of attacks.

Internet is a pool of information, which contains billions of text documents which are stored in compressed format. In literature we can find many text classification algorithms which work on uncompressed text documents. In this paper, we propose a novel representation scheme for a given text document using compression technique. Further, proposed representation scheme is used to develop a methodology to classify the text documents. For the purpose of representation, we have used LZW compression technique and the dictionary representation obtained by LZW technique is used as representative for the text document. For classification we have used nearest neighbor method. Extensive experimentation is carried out on seven datasets, out of which three are our own datasets and remaining four are publically available datasets resulting with approximately 80% of F-measure.

This paper introduces video shot detection method based on chromaticity histogram. The Chromaticity diagram provides a two dimensional representation of the image, and a corresponding two dimensional histogram can be constructed. The Horizontal Projection Profile (HPP) and Vertical Projection Profile (VPP) are obtained from the histogram. This feature vector is compared between successive frames to detect the presence of cut in the video. Experiments have been conducted on TRECVID video database to evaluate the effectiveness of the proposed model.

The ever increasing amount of XML documents available on the World Wide Web demands automated tools and techniques that would make the search and retrieval of XML documents more effective and efficient. Classification of XML documents is one of the significant tasks which are being explored by many researchers in this direction. Due to the presence of inherent structure in the XML documents, conventional text classification methods cannot be used to classify XML documents directly. Hence, there is a need for the development of tools and techniques that automatically classifies XML documents. In this work, we have developed an algorithm based on ‘k’ nearest neighbors to classify XML documents by considering both the content and structure. The developed algorithm is tested on a subset of MEDLINE dataset for different values of ‘k’ and varying size of training set and the results are tabulated.