Akshay Girdhar

Multimodal Biometric System Using Speech and Signature Modalities

This paper discusses about Multimodal Biometric System which are used to overcome some of the problems of unimodal systems like noise in sensed data, intra-class variations, distinctiveness, and spoof attacks. Multimodal biometrics is the combination of two or more modalities such as signature and speech modalities. In this work an online signature verification system and speaker verification system are combined as these modalities are widely accepted and natural to produce. Although this combination of multimodal enhances security and accuracy, yet the complexity of the system increases due to increased number of features extracted out of the multiple samples and suffers from additional cost in terms of acquisition time. So these days the key issue is at what degree features are to be extracted and how the cost factor can be minimized, as the number of features increases the variability of the intra-personal samples due to greater lag times in between consecutive acquisitions of the sample also increases. Increase in variability of the system will further increase FAR. Thus to resolve these issues an effective fusion level and fusion mode is required. This paper presents a novel user authentication system based on a combined acquisition of online pen and speech signals.

Vehicle Number Plate Detection System for Indian Vehicles

An exponential increase in number of vehicles necessitates the use of automated systems to maintain vehicle information. The information is highly required for both management of traffic as well as reduction of crime. Number plate recognition is an effective way for automatic vehicle identification. Some of the existing algorithms based on the principle of learning takes a lot of time and expertise before delivering satisfactory results but even then lacks in accuracy. In the proposed algorithm an efficient method for recognition for Indian vehicle number plates has been devised. The algorithm aims at addressing the problems of scaling and recognition of position of characters with a good accuracy rate of 98.07%.

Enhanced Iris Recognition System - an Integrated Approach to Person Identification

This paper discusses about Enhanced iris recognition which is used to overcome some of the problem like to automate the recognition of the iris by reducing complexity and increasing algorithm speed. Various challenges are faced while working with the iris recognition system. Iris recognition systems make use of the uniqueness of the iris patterns to derive a unique mapping. Iris recognition, as a biometric method, outperforms others because of its high accuracy. Iris recognition also has the ability to handle very large populations at high speed. Mostly three stages are followed while working with iris system i.e. preprocessing, feature extraction and recognition stage. This paper presents an automated and novel iris recognition system where overall computational match speed is reduced (from iris preprocessing to the final stage of recognition) and hence makes system more reliable with accuracy of 99.38% and low FAR.

Region Based Adaptive Contrast Enhancement of Medical Ultrasound Images

Since the dawn of computerization of medical science, ultrasonic imaging has been an active research area and impacts important applications in the medical field. Ultrasound images have low contrast due to various artifacts. In most of the existing region growing contrast enhancement techniques, selection of seed point and threshold value is a challenging task. In this paper, an adaptive region based contrast enhancement technique based on the region growing segmentation concept is proposed. In the proposed work, automatic selection of seed point from the region of interest of an image is done followed by computation of threshold value to segment foreground and background pixels where the prime focus is to enhance foreground areas of an image. Qualitative and quantitative evaluation of the results of the proposed technique has been performed and results are very promising when compared with state-of-the-art techniques.

Region of Interest Based Contrast Enhancement Techniques for CT Images

Medical imaging refers to an approach through which radiographers diagnose the human body using X-rays, CT, ultrasound, and magnetic resonance techniques. It is widely used safe imaging system in medical field that assists in extracting and visualizing the fine details from the image. In order to extract the details from an image, it should be of high quality, but presence of noise in images makes the image unclear and due to this the image enhancement is referred to as the most significant matter of concern in image processing. Every time when the patient is diagnosed, who is suffering from one or the other kind of disease, using techniques of medical imaging, a tumour is located in their body. This tumour can be called as the affected area in the patients body. To locate this region of interest, various techniques like Otsu method are already available. The intensity difference between the tumour and the other body parts is either low or high but the major problem in detecting the tumour arises when the intensity difference between the tumour and the other organs is very low. Therefore, in order to extract the region, there is a need to enhance the contrast of affected region, so as to distinguish them from other parts. Region of interest based contrast enhancement is a technique to generate and extract the region based on the image contrast of the image. The primary objective of the work is to design a novel algorithm for extracting the affected region. A technique based on level set evolution is described for extracting the region. The proposed technique provides better image quality and contrast to noise ratio. The main focus in the proposed work is on contrast enhancement of CT images based on un-sharp mask filter as a necessary pre-processing. The results thus obtained have been compared with the existing state-of-the-art techniques and the outcomes show that the results are very promising.

Virtualization Approach to Cluster Based Winograd's Variant of Strassen's Method Using RMI

Virtualization helps a lot in handling critical applications in a highly availability manner, stream lining in applications deployment, application migrations etc. It is because of the virtualization that users are able to reduce the extra requirements of hardware, power consumption, trimming the space requirements, air conditioning requirements etc. Server consolidation, disaster recovery, dynamic load balancing, testing and deployment, virtual desktops, improved system reliability all these facilities are being provided by the virtualization techniques. Our given proposal of work uses the same for implementing cluster programming in order to implement Winograds variant of Strassens matrix multiplication. An API known as RMI (Remote Method Invocation) is being used with the help of which programmer can create distributed applications so that objects which are residing on the different systems can interact with each other in an efficient manner. The divide and conquer approach of the Winograds variant method is the main factor which helps us in distributed computing. Partitioning a given matrix into sub matrix is done at the master side and at each slave the logical partitioning into 2*2 matrixes has been done for implementing given algorithm. For the purpose of analysis of the given proposed work various performance metrics like excessive parallel overhead speed up, efficiency, and total execution time are being used.

RMI Approach to Cluster Based Cache Oblivious Peano Curves

There are number of problems that are so complex/large that it becomes impractical or even in some cases impossible to solve these problems on a single machine. As compared to the serial computation, parallel computation is much result oriented for understanding, simulating of number of complex and real world physical process. The cache oblivious(CO) model helps us in designing the algorithms which are cache alert. Moreover these algorithms will be independent of the given systems cache size. A matrix multiplication based upon the Peano curves helps in designing of the cache oblivious algorithms. The distributed environment is being developed using RMI (Remote Method Invocation). In this setup the Master system will decompose a large size matrix into the smaller (ones depending upon the system available). The slave systems will perform the computations as per the equations based upon space filling Peano curves which are cache oblivious in nature. As a result we are able to reuse the matrix elements again and again which leads to decrease in number of cache misses and increasing the overall execution time of whole cluster. At the master system actual partitioning is done to generate submatrix and the virtual partitioning into size of 3x3 is being done at the slave systems for implementing multiplication based upon Peano curves(PC). PC algorithmic approach provides spatial locality which is a basic requirement for increasing the overall system efficiency.

A New Thresholding Technique for Despeckling of Medical Ultrasound Images

Ultrasound imaging is the most commonly used imaging system in the medical field. Due to various sources of interferences medical images are often deteriorated by noise. Main problem to this imaging technique is introduction of speckle noise. Speckle noise is a mottling of the image with dark and bright spots, which degrades the fine details and quality of image. The primary objective of this paper is to implement a multi-resolution based method for despeckling of ultrasound images and compare it with existing state-of-art techniques. There have been several methods for efficient reduction of speckle noise present in ultrasound images in order to reduce the noise level and improve the visual quality for better diagnoses. In this paper an improved method is given which can determine an optimal threshold and neighbouring window size for every subband by the Steins unbiased risk estimate (SURE). Experiments demonstrate that the proposed method achieves better results than some of the best adaptive and non-adaptive despeckling techniques. The proposed method can also be extended to the dual-tree complex wavelet transform (DT-CWT).

Intrusion Response System in AODV for MANET

The mobile adhoc network (MANET) requires effective intrusion response system. In this paper, we present an intrusion response system that supports the infrastructureless nature of MANETs. We propose a NHELLO and Link Layer based solution towards excluding malicious node which is robust against address spoofing from the attacker. In particular, we investigate how power adaption can be used to keep a malicious node away from normal node’s transmission range. Important issue in this strategy is to select optimal transmission power so that malicious node goes out of operating zone of network, as well as node adapting power itself remains in the operating zone. We also provide a detailed performance evaluation based on various network parameters i.e. a series of simulation studies. Our results show that the proposed concept significantly improves the overall security of mobile ad hoc network without having geographical information of nodes.

Computer Vision-Based Tomato Grading and Sorting

Since ages, agricultural sector plays an important role in the economic development of a country. In recent years, industries have started using automated systems instead of manual techniques for quality evaluation. In agriculture field, grading is very necessary to increase the productivity of the vegetable products. Everyday a huge amount of vegetables are exported to other places and earn a good profit. So, quality evaluation is important in terms of improving the quality of vegetables and gaining profit. Traditionally, the vegetable grading and classification were done through manual procedures which were error prone and costly. Computer vision-based systems provide us such accurate and reliable results that are not possible with human graders/experts. This paper presents a vegetable grading and sorting system based on computer vision and image processing. For this work, tomatoes have been used as a sample vegetable. A total of 53 images were acquired using own camera setup. Afterward, segmentation using Otsu’s method was performed so as to separate the vegetable from the background. The segmented images, thus obtained, were used to extract color and shape features. At last, grading and sorting were performed using backpropagation neural network. The proposed method has shown an accuracy of 92% and outperformed the existing system.