A. Annis Fathima

Multiclass object detection system in imaging sensor network using Haar-like features and Joint-Boosting algorithm

This paper proposes an efficient scheme for detecting different object classes in an imaging sensor network. The object detection system detects all the instances of objects (for which the classifier was trained) in the given image, regardless of their scales and locations. Therefore, the image can be thus seen as a set of sub-windows that are to be evaluated by the detector. The detector selects those sub-windows that contain the instances of the objects trained. The traditional approach for multiclass object detection is to use different independent classifiers to the image, at multiple locations and scales. This can be slow and requires a lot of training data. To achieve a fast and robust implementation, shared features are used. In the existing schemes, part-based models have been used for evaluating the object features, so these features being more class-specific cannot share more information among different classes. Hence in this paper, rectangular features called Haar-like features which are more generic is used and thus more number of features can be shared. The proposed scheme uses Joint-Boosting algorithm for training the multiclass object classifier. The benefits of illumination normalisation or variance normalisation technique used to neutralise the effect of changing lighting conditions are explored. Though the proposed scheme is validated for car and pedestrian classes, the training and detection techniques used in this scheme can be generalised for any object class.

Image Stitching with Combined Moment Invariants and Sift Features.

Abstract Image stitching is used to combine multiple photographic images from camera network with overlapping field of view to produce panoramic view. With image stitching, the view is enlarged and the amount of information increases with the no. of images that are stitched. In the existing methods, the whole images from the adjacent views are considered thus leads to increase in both time and computational complexity. In this paper, an approach for image stitching using invariant moments combined with SIFT features is presented to reduce the time and computational complexity. It is observed that only a small portion of the adjacent view images are overlapped. Hence, the proposed method aims in detecting overlapping portion for extracting matching points. The overlapping regions are determined using gradient based dominant edge extraction and invariant moments. In the deduced region, the SIFT (Shift Invariant Feature Transform) features are extracted to determine the matching features. The registration is carried on with RANSAC (Random Sample Consensus) algorithm and final output mosaic is obtained by warping the images. The proposed approach results in reduced time and computational when compared to existing methods.

An Efficient Face Recognition System Using DWT-ICA Features

Multiresolution representations and Subspace analysis have been widely accepted in the face recognition systems. This research paper combines the benefits and presents the feature extraction method using Discrete Wavelet Transform (DWT) and Independent Component Analysis (ICA). The DWT provides multiresolution representations and are effective in analyzing the information content of the image and generates the feature sets for images from individual wavelet sub bands. The feature images constructed from Wavelet Coefficients (Cohen Daubechies Feauveau (CDF-9/7)) are used as a feature vector for ICA based subspace analysis. ICA is an unsupervised statistical method reduces the dimensionality of the feature vector and extracts the information in the higher-order relationship of pixels. ICA method has been used to find statistically independent basis images or coefficients for the face images to deal with the sensitivity to higher order image statistics. Reduced feature vector are used for further classification using Euclidean Distance (ED) classifier. The proposed scheme has been tested on the standard and real-time Database and the results have been reported. It was observed that the proposed method classifies the images with better accuracy and outperforms the existing methods.

Object Detection and Tracking in Secured Area with Wireless and Multimedia Sensor Network

This paper presents a scheme for object detection and tracking in heterogeneous sensor network environment. The main objective is to provide a solution based on Wireless and Multimedia Sensor Networks (W&MSN) for monitoring and tracking of object (person/vehicle) in secured area. The multi-tier, heterogeneous sensor network adapted for efficient usage of image data. The object detection is carried out with background subtraction technique. The detected blob region is taken as input for extracting the features based on Haar wavelet. The feature extraction is followed by joint boosting algorithm to classify as interested object or not. The object detection is combined with Kalman Filter to accurately track the movement of desired objects in the given area. This approach provides better detection and tracking of person even in the presence of occlusion and multiple persons in the environment.

Mobile Authentication Using Iris Biometrics

In today’s fast moving world, mobile phones have become one of the basic needs and mobile security is of a major concern. Mobile security is needed to assure a secured method for mobile transactions and to preserve data integrity and confidentiality. The present method of security involves password authentication. However this method is highly vulnerable to spoof attacks. Biometrics based authentication is a popular method of providing security. This paper proposes a novel method to provide security in mobile phones using biometrics. Among all the biometric modalities, Iris is proven to be one of the best traits and most suitable for authenticating mobile phone users. The challenging issue in the iris based authentication is localizing iris, the Region of Interest (ROI) and extracting features for real-time images due to varying illumination conditions. The proposed scheme adapts Sobel operator in color space and Contour method to accurately detect and segment the iris from eye image. The feature extraction is by Discrete Wavelet Transform (DWT), for accurate classification, simple k-Nearest Neighbor (k-NN) is taken and based on the percentage of match the authentication is done. The proposed algorithm is using JavaCV (Java + OpenCV), tested in Android 2.2 platform and implemented in Samsung I9003 Galaxy S with Android 2.2 OS, processing speed of 1 GHz and Internal Memory of 4GB.

Face recognition system using Combined Gabor Wavelet and DCT approach

In this paper, an approach for face recognition combining multi-resolution analysis and transform domain analysis is proposed. Face Recognition system find its use in many applications such as authentication, surveillance, human-computer interaction systems etc. As the applications using face recognition systems are of much importance and demand more accuracy, more robustness in the face recognition system is expected with less computation time. In the proposed ComGW-DCT approach, features are extracted using a combination of Gabor filters and Discrete Cosine Transform (DCT). The normalised input grayscale image is approximated and reduced in dimension to lower the processing overhead for Gabor filters. This image is convolved with bank of Gabor filters with varying scales and orientations. Further DCT technique is adapted to reduce the feature space dimension. DCT extracts low frequency components of the Gabor wavelet thus resulting in the compression of Gabor features. For classification, k-Nearest Neighbour (k-NN) classifier is used to recognise the test image by comparing with each of the training set features. The ComGW-DCT approach is robust against illumination conditions as the Gabor features are illumination invariant. This algorithm also aims at better recognition rate using less number of features for varying expressions without affecting the computation time. The results of the proposed system are evaluated using AT&T database and MIT-India face database.

Performance analysis of multiclass object detection using SVM classifier

Multiclass object detection is considered for detecting different object classes in a cluttered environment. Traditional approaches require applying a battery of different classifiers to the image with a large number of complex features used to detect the objects. Specialized detectors usually excel in performance, while the class-specific features increase detection accuracy, but at the expense of complexity. In this paper, an efficient method of human face and car detection using cascaded structure of independent object classifiers is proposed. The approach is based on background elimination using statistical features, followed by foreground detection using Principal component analysis (PCA) and Histogram of Gradients (HoG) with SVM classifier. For detecting the object of interest from the image, the system primarily filters the potential object area by analyzing the local histogram distribution. After background elimination, the trained classifier detects foreground using higher order parameters like PCA for human faces and HOG for cars. In this paper, the kernel function for SVM classifier, suitable for individual object classifier is analysed based upon ROC-AUC parameter. The proposed system is implemented in Matlab. The system is validated with performance metrics like precision, recall and accuracy.

Decision Level Fusion Framework for Face Authentication System

In this paper, multiple algorithm and score-level fusion for enhancing the performance of the face based biometric person authentication system is proposed. Though many algorithms are conferred, several crucial issues are still involved in the face authentication. Most traditional algorithms are based on certain assumptions failing which the system will not give appropriate results. Due to the inherent variations in face with time and space, it is a big challenge to formulate a single algorithm based on the face biometric that works well under all variations. This paper addresses the problem of illumination and pose variations, by using three different algorithms for face recognition: Block Independent Component Analysis (B-ICA), Discrete Cosine Transform (DCT) and Kalman filter. The weighted average based score level fusion is performed to improve the results obtained by the system. An intensive analysis of the various algorithms has been performed and the results indicate an increase in accuracy of the proposed system.

Hybrid approach for face recognition combining Gabor Wavelet and Linear Discriminant Analysis

Face Recognition system finds many applications in surveillance and human computer interaction systems. As the applications using face recognition systems are of much importance and demand more accuracy, more robustness in the face recognition system is expected with less computation time. In this paper, a Hybrid approach for face recognition combining Gabor Wavelet and Linear Discriminant Analysis (HGWLDA) is proposed. The normalized input grayscale image is approximated and reduced in dimension to lower the processing overhead for Gabor filters. This image is convolved with bank of Gabor filters with varying scales and orientations. LDA, a subspace analysis techniques are used to reduce the intra-class space and maximize the inter-class space. The techniques used are 2-dimensional Linear Discriminant Analysis (2D-LDA), 2-dimensional bidirectional LDA ((2D)2LDA), Weighted 2-dimensional bidirectional Linear Discriminant Analysis (Wt (2D)2 LDA). LDA reduces the feature dimension by extracting the features with greater variance. k-NearestNeighbour (k-NN) classifier is used to classify and recognize the test image by comparing its feature with each of the training set features. The HGWLDA approach is robust against illumination conditions as the Gabor features are illumination invariant. This approach also aims at a better recognition rate using less number of features for varying expressions. The performance of the proposed HGWLDA approaches is evaluated using AT&T database, MIT-India face database and faces94 database. It is found that the proposed HGWLDA approach provides better results than the existing Gabor approach.

Multi-class object detection by part based approach

This paper presents an efficient method to detect multiple objects in multiple views by part based approach in computer vision. The part based method is adapted to detect and classify the multiple parts of objects as car/person in order to overcome the occlusion. For detecting the multiple instances of object, the cascaded structure is considered, with each node as joint boosting classifier with shared features. Features extracted are Haar-rectangular features, as it efficiently captures the structural property of the object. With joint boosting algorithm, the features are shared among different classes, thus in turn reducing the computational complexity and detection time. The classifier efficiency is analysed by two parameters namely precision and recall. Although the proposed scheme is validated for car and pedestrian classes, the training and detection techniques used in this scheme can be generalized for any object class.