Digital Connectivity – Social Impact

This paper presents a novel approach for texture classification using Shearlet Transform. The Shearlet Transform is a recently developed tool, which have the multiscale framework which allows to efficiently encode anisotropic features in multivariate problem classes. Shearlets are a newly developed extension of wavelets that are better suited to image characterization. In addition the degree of computational complexity of many proposed texture measures are very high. In this paper, a novel texture classification method that models the adjacent shearlet subband dependences. In this paper the classification efficiencies of Minimum Distance classifier was compared with SVM classifier efficiency. For texture classification, the energy features are used to represent each shearlet subband. Comprehensive validation experiments performed on different datasets proves that this research work outperforms the current methods due to efficient multiscale directional representation of Shearlet Transform.

With the increase in use of e-technologies, large amount of digital data are available on-line. These data are used by both internal and external sources for analysis and research. This digital data contain sensitive and personal information about the entities on which the data are collected. Due to this sensitive nature of such information, it needs some privacy preservation procedure to be applied before releasing the data to third parties. The privacy preservation should be applied on the data such that its utility during data mining does not get reduced. ℓ-Diversity is an anonymization algorithm that can be applied on dataset with one sensitive attribute. Real life data contain numerous sensitive attributes that have to be privacy preserved before publishing it for research. This paper proposes an Enhanced ℓ-diversity algorithm that can diversify multiple sensitive attributes without partitioning the dataset. Two datasets namely, bench mark Adult dataset and Real life Medical dataset are used for experimentation in this work. The privacy preserved datasets using the proposed algorithm are compared for its utility with ℓ-diversified dataset for single sensitive attribute and original dataset. The results show that the proposed algorithm privacy preserved datasets have good utility on selected classification algorithms taken for study.

Pictures are considered to be the capture of a real world scenario and we second that “pictures speak more than words”. It is also understandable that information can also be taken out of an image. To extract the information out of the image, the most important feature is the edge. There are many techniques to detect edges. This paper has a detailed study about these different techniques and proposed an intelligent algorithm. Also an analysis of these algorithms on various datasets are done.

Malware holds an important place in system performance degradation and information embezzling from the victim system. Most of the malware writers choose their path to reach the victim system through the internet, infected browsers, injected files, memory devices, etc., highly obscured malwares evade the automated tools installed in the victim. Once the victim system gets affected by the malware, executable processes are controlled by malware. In this paper, an algorithm has been developed to identify the malware using image processing. The malware detection process has three phases. In first phase, the files (.exe) are converted into a gray scale image. The binary values of corresponding files are converted into 8 - bit gray scale intensity value. The band pass frequency of gray scale image is computed in second phase. In the final phase, third and fourth order statistical parameter such as skewness and kurtosis are calculated at the each sub region of band pass frequency image. The region which has the highest skewness and kurtosis value is marked as the malware file. The detection performance of the proposed method has been evaluated by using 1300 portable executable files. The detection method has a true positive ratio of 93.33% with 0.1 false positives. Preliminary results indicate that the proposed algorithm is better than other conventional malware detection methods.

The efficiency of an iris authentication system depends on the quality of the iris image. Denoising of the iris image is indispensable to get a noise free image. In this paper, a novel method is proposed to remove Gaussian noise present in the iris image using Undecimated wavelet, a threshold based on Golden Ratio and weighted median. First, decompose the input image using Stationary Wavelet Transform (SWT) and apply the modified Visushrink to the wavelet coefficients using hard and soft thresholding. Then apply inverse SWT to get the noise free image. Different kinds of wavelet filters such as db1, db2, sym2, sym4, coif2 and coif4 for different noise levels are performed. The filter db1 is outperformed. In this research, experiments have been conducted on the iris database CASIA. The Peak Signal-to-Noise Ratio (PSNR), Signal-to-Noise Ratio (SNR), Root Mean Square Error (RMSE) and Mean Square Error (MSE) have been computed and compared.

Stemming is the process of generating root word from the given inflectional word. Tamil Language has technical challenges in stemming because it has rich morphological patterns than other languages, so Analogy Removal Stemmer (ARS) is proposed in this research, to find stem word for the given inflection Tamil word from text corpora. The performance of the proposed approach is compared with Light Stemmer (LS) and Improved Light Stemmer (ILS) algorithms based on correctly and incorrectly predicted stem words. The experimental result clearly shows that the proposed approach ARS for Tamil corpora performs better than the LS and ILS algorithm.

Recommender systems serve as business tools which make use of knowledge discovery techniques to reshape the world of E-Commerce. Collaborative filtering (CF), the most effective type of recommender systems, predicts user preferences by learning from past user-item relationships. Prediction algorithms are based on similarity between item vectors or user profiles. However similarity computations become less efficient if item vectors or user profiles do not contain enough ratings. A technique which is based on Pseudo Relevance Feedback is proposed to expand item vectors in order to make them contain more ratings. The proposed approach first expands item profiles and refines the expansion in order to remove expansion deviations. The experiments on MovieLens data set show that the proposed technique is efficient in expanding the rating matrix and outperforms state of the art collaborative filtering techniques for providing more efficient predictions.

Service Discovery is a process of identifying the compatible resources of a transaction to result it in a comfort zone. The environment of Internet of Things would be a mixture of hardware and software independent components. Offering a service in an IoT architecture is challenging problem based on participant registration and logs. A model for Integrating Physical world devices in constrained web environments using Constrained Application Protocol (CoAP) together with an end-to-end IP and RESTful Web Services based architecture is proposed. We suggest a Visa Processing Algorithm for an IoT environment to maintain time constraint participants and for anonymity finding. The proposed model is an integrating Constrained Application Protocol with the algorithm for service discovery issues particularly handling anonymous entity.

Preprocessing is one of the steps in Data Mining, which involves Noise removal, Identification of outlier, Normalization, Data transformation, Handling missing values, etc. Missing value is a common problem in large datasets. Most frequently used method to handle missing values by statistical is discarding the instances with missing values. Sometime deletion of instances with missing values cause loss of essential information, which affects the performance of statistical and machine learning algorithms. This paper focuses on handling missing values using unsupervised learning techniques. Rough K-Means based missing value imputation was proposed and compared with K-Means, Fuzzy C-Means based imputation methods. The experimental analysis is carried out on two data sets Lung Cancer and Cleveland Heart data sets. The proposed method achieves the best accuracy for some of the datasets.

This paper focuses to address the suppression of residual speckle content, thereby leading to performance improvement of Anisotropic Diffusion (AD) filtering. The present work proposes an optimized AD filtering approach for despeckling of Synthetic Aperture Radar (SAR) images using Ant Colony Optimization (ACO). The residual speckle suppression has been attained via optimal parameter selection of parameters of AD using ACO algorithm. Further, computation of conductance function via eight-directional gradients (derivatives) leads to effective edge preservation during despeckling. During simulations, the fidelity of restored SAR images is validated using PSNR and SSIM as image quality metrics.

Women Breast Cancer has high incidence rate in worldwide. Computer aided diagnosis helps the radiologist to diagnose and treat the breast cancer at early stage. Recent studies states that Adaptive Neural Fuzzy Inference System (ANFIS) classifier achieves notable performance than the other classifiers. The major forte of ANFIS is that it has the robust learning mechanism with fuzzy data. However, the connections between the layers are not pruned for their significance. An Ant Colony Optimization (ACO) based learning is proposed in this paper to improve ANFIS classifier with a novel pruning strategy. The proposed algorithm is inspired from the social life of a special species called ‘Weaver Ants’. The proposed classifier is evaluated with the mammogram images from MIAS database, the quantified results show that this weaver ant based learning strategy improves the ANFIS classifier’s performance.

A Wireless sensor network (WSN) is composed of a large number of sensor nodes, which perform multiple tasks, namely sensing, data processing and forwarding of observed data. WSNs nodes may possess sensitive data that are prone to various attacks. For such a network to be viable, integrity and authenticity should be provided to the data generated by the sensor nodes. For example, in military surveillance and enemy tracking applications, the localization system of the nodes is the target for many attackers. In such applications, the base station would broadcast the command for localization to all the sensor nodes in the field. The sensor nodes would respond to this query with the required data. Any compromised node at this point would generate false data and may lead to miscalculation of the localization process and incorrect decision making. Hence, a resilient authentication is necessary to authenticate a node. As a first step towards this objective, a lightweight identity based signature for authentication of the sensor nodes is proposed in this paper. The scheme uses “fingerprint”, i.e. a lifetime secure memory fraction in the sensor nodes as a parameter for signature generation. In addition, the parameters for fingerprints are generated dynamically and the computed fingerprint values are not stored permanently in the hardware. Because of these features, the sensor nodes can overcome identity based attack like Sybil attack. Also, it is impossible to read the contents of the sensor node even if the node is captured by the attacker. The security proof for this scheme is based on the Computational Diffie-Hellman assumption and proved in the random oracle model. On the computation point of view, the proposed scheme requires minimal operations in signing than the existing identity based signature approaches.

A mobile ad hoc network (MANET) is a type of infrastructure-less network with mobile nodes communicating with each other. The distributed administration and dynamic nature of MANET makes it vulnerable to variety of security attacks. So there is a need to secure the data transmission. The Optimized link state routing protocol is an efficient proactive routing protocol which is suitable for such dense and large scale MANET. In this paper, secure-OLSR (sOLSR) is proposed to ensure the secure path data transmission. A new algorithm has been proposed which calculates the trust values for each node. Based on the trust values, the malicious nodes are found and then a path is selected from the available paths based on the maximum path trust. Simulation results using NS-3 simulator shows that the proposed mechanism is very effective than the existing trust based protocols in terms of packet drop ratio, average latency and overhead.

Live virtual machine (VM) migration relocates running virtual machine from source physical server to the destination physical server without compromising the availability of service to the users. Live VM Migration guarantees energy saving, fault tolerance and uninterrupted server maintenance for the cloud datacenter. The workload handled by the cloud datacenters are unpredictable in nature. Hence, the migration needs intense planning. Resource starvation occurs due to dynamic nature of workload handled by cloud datacenter. The objective of this paper is to predict the resource requirement of the virtual machines running various workloads and to appropriately place them during migration. The resource requirement of the running virtual machines are predicted using combined forecast technique. The combined forecasting technique improves the forecasting accuracy. Every host machine suitably migrates based on the current and forecasted utilization. The proposed algorithm has been validated using set of simulations conducted on Google Datacenter Traces. The results show that the proposed methodology improves the forecasting accuracy.

In IT services SOA is one of the most elastic and modular approaches and it is a prerequisite for arising technologies like cloud, these cloud services are exposed as web services based on industry standards, which follows WSDL for service illustration. These services depend on SOAP to handle service request and response. Hence Web services security is one of the important factors which are used to assess cloud system security. While creating a new web service or with an existing web service communication, it is prudent to have secure data transmission with end users identity such as card numbers, user names, passwords etc. Security standards like WS-Security only addresses message integrity, confidentiality, user authentication, and authorization. The proposed system offers confidentiality and integrity protection from the creation of the message to its consumption. This system will look at a color palette scheme which records the RGB color values of the chosen color during registration and these values are used during sign on, subsequently it performs the access control mechanism. To strengthen web services towards message level by encrypting SOAP messages with AES and shared key is derived using new cryptosystem called Rbits (Random bits) cipher as a service and digital signature handler facilitates secure key exchange which is completed ahead with SOAP message generation. The essential aspect of this proposed system is from core key form multiple random keys which safeguards the messages with highest possible immunity to crack when the applications or services communicating with web services.

A MANET (Mobile Ad-Hoc Network) is composed of mobile, autonomous, wireless nodes that could be connected at network edges to that of fixed wired internet. The ad-hoc network does not need infrastructures that are required for other wireless networks. The term infrastructure includes need of base stations, routers etc. The network execution ought to be supported up by protecting collaboration among various nodes and it is finished by essential mechanisms. In further, most proficient routing protocols are utilized to route the packets to the destination from the source by the method of routing process. In MANET, there are various kinds of routing protocols and every one of them is associated with systematic circumstances. This paper focuses on designing a new routing protocol which considering minimum number of nodes for routing and select optimal route. This protocol also concentrates clustering as well as security considerations in order to provide tenable and proficiency routing.

Social networks have acquired much attention recently, largely due to the success of online social networking sites and media sharing sites. In such networks, rigorous and complex interactions occur among numerous one-of-a-kind entities, main to massive statistics networks with notable enterprise capacity. Community detection is an unsupervised learning task that determines the community groups based on common interests, occupation, modules and their hierarchical organization, using the information encoded in the graph topology. Finding communities from the social network is a difficult task because of its topology and overlapping of different communities. In this research, the Girvan-Newman algorithm based on Edge-Betweenness Modularity and Link Analysis (EBMLA) is used for detecting communities in networks with node attributes. The twitter data of the well-known cricket player is used right here and community of friends and fans is analyzed based on three exclusive centrality measures together with a degree, betweenness, and closeness centrality. Also, the strength of extracted communities is evaluated based on modularity score using proposed method and the experiment results confirmed that the cricket player’s network is dense.

GIS based Smart Energy Infrastructure Architecture and Revenue Administration is an integrated framework of web, mobile and GIS technology to manage electrical infrastructure and produce energy bills for the consumers. This framework helps to plan new electrical transmission infrastructure needed for quality power supply and assist in detailed planning for infrastructure. This helps to create state-of-the-art Geomatics oriented models which can assist in decentralized planning and development for robust development. The Revenue Administration Model being proposed is easily adaptable which allows for ultimate flexibility as government processes may change over a time due to changes in tariff, as well as being able to easily integrate with external applications such as Revenue accounting, ATP, IVRS, GIS, Spot billing, Payment Gateway, SMS alert and work management. The framework focuses on practical steps to be carried out for exploiting full potential of technology convergence, with emphasis on technically viable smart energy infrastructure keeping in view of sustainable growth.

Radio Frequency Identification (RFID) is a valuable technology for tracking objects in the supply chain. Security and privacy requirements arise with the fast deployment of RFID in supply chain in a heterogeneous environment. Authentication is one of the important security requirements in cloud environment. Even though several RFID cloud authentication protocols are available for supply chain management, they lack to satisfy some security requirements. There is a need for secure, efficient, and scalable protocol for agile supply chain. In this paper, an RFID cloud authentication protocol is proposed and an informal security analysis is carried out. Performance analysis is done with respect to the tag entity. The proposed protocol is scalable and it preserves tag/reader privacy, provides mutual authentication and resistant to many attacks. Comparison with the existing protocol in terms of communication cost shows that our protocol outperforms the other protocols.

Automatic segmentation of continuous speech plays an important role in building promising acoustic models for a standard continuous speech recognition system. This needs a lot of segmented data which is rarely available for many languages. As there are no industry standard speech segmentation tools for Indian languages like Tamil, there arises a need to work on Tamil speech segmentation. Here, a segmentation algorithm that is based on Graph cut is proposed for automatic phonetic level segmentation of continuous speech. Using graph cut for speech segmentation allows viewing speech globally rather locally which helps in segmentation of vocabulary, speaker independent speech. The input speech is represented as a graph and the proposed algorithm is applied on it. Experiments on the speech database comprising utterances of various speakers shows the proposed method outperforms the existing methods Blind Segmentation using Non-Linear Filtering and Non-Uniform Segmentation using Discrete Wavelet Transform.

Diabetic Retinopathy (DR) is the root cause for retinal blood vessel damages among the diabetic patients. If it is not identified and treated earlier, at the later stage it leads to 100% vision loss. Thus there is a need of a system to identify the early stage of DR, so that it can be treated according to ETDRS (Early Treatment Diabetic Retinopathy Study). The proposed Pulse Coupled Neural Network (PCNN) model segments the retinal blood vessels from the depigmented fundus images and provides the structure of the retinal blood vessels. This segmented blood vessel map helps the ophthalmologist to identify the severity level of the blood vessel damages and to treat the early Diabetic Retinopathy among different age group populations. The proposed PCNN model is applied over the DRIVE database and the results are compared with various supervised and unsupervised segmentation approaches. The proposed method improves the accuracy in detecting the tiny blood vessels in the depigmented fundus images than other existing methods. This system increases the number of true positives; true negatives and reduces the false positives, false negatives while compared with the ground truth images. The Specificity of the proposed system over DRIVE database is 99.31%, Sensitivity is 67.54% and Accuracy is 97.23%. The resultant image of the segmented blood vessels can be used for further diagnosis and to measure the severity level of DR.

Bones that provide the structural support of the body, are composed of many inorganic compounds and organic materials that all together can be used to determine the mineral density of the bone. The bone mineral density (BMD) is an index measure that is widely used as an indicator of the health of the bone. A densitometry study from dual X-ray absorptiometry (DEXA) system is a popularly used method to assess BMD. BMD values vary depending on race, age, gender and other health conditions. As DEXA is quite an expensive method and requires frequent calibration process to work properly, in this paper, we explore the possibility of developing an affordable and reliable system depending on single X-ray absorptiometry with the use of supervised learning methods. The methodology based on inference is tested on a data set consisting of spine and pelvis X-ray images of patients of varying ages between 10 and 90 years of PSG Hospitals, Coimbatore, India and the results proved to be an indicator of density of the bone.