Subhash Kulkarni

Skeleton based Signatures for Content based Image Retrieval

Content Based Image Retrieval with fast and high matching retrieving ability is the need of the day for shape mining. A simple, fast, robust, invariant and efficient Content Based image Retrieval system with shape signatures derived from skeleton, region and boundary of the object is presented. The shape signatures derived from distance mapped function are invariant to rotation and scaling.

Novel Ensemble Neural Network Models for better Prediction using Variable Input Approach

In this work, seven Ensemble Artificial Neural Network (ANN) models, namely, Multilayer Perceptron Network (MLPN), Elman Recurrent Neural Network (ERNN), Radial Basis Function Network (RBFN), Hopfield Model (HFM), Ensemble Neural Network based on Variable Inputs with No Hidden Layers (ENN-V-S), Ensemble Neural Network based on Variable Inputs with Hidden Layers (ENN-V-M), Ensemble Neural Network based on Time Inputs with No Hidden Layers (ENN-T-M), are developed to predict the rainfall for one of the large cities of India i.e. Bangalore. Different network models are developed to match the predicted results with the actual data and ENN-Average is found to be the best among all. In order to test this, actual rainfall data was collected in Bangalore city for the calendar years 2007, 2008 and 2009. This data was used as training data for the ANNs and predictions were made for the year 2010. Again these predictions were compared with the actual data to verify the performance of the ANNs. In this study, it has been proved that ENN-Average model based on back propagation algorithm provide better accurate predictions than the SNN and ENN models based on other algorithms.

Handover Modeling of Multiple States of Mobile Node in a Five Node Network Model

need to be assessed to identify the one which produces best results in terms of successful handover. The performance measuring algorithms should be independent of the technology used in the handover algorithm. The performance can be based on number of successful handovers or number of incorrect decisions. Hence it necessitates a common algorithm to measure the performance. There are some performance measurement algorithms available in [4, 5]. These algorithms do not have robustness in terms of analytical approach and mathematical formulation. However this gap was filled by a new approach [6] based on the probability models and is independent of the technology used for the handover. The approach is known as wrong decision probability and is based on number of incorrect decisions made for hand over. Wrong decision probability approach is based on the criteria like bandwidth availability, signal strength, movement of the mobile node etc. Authors in [6] used a two network model which was academic in nature. However, in practice the number of networks are more than two, many cases it is five. Suresh et. al. developed a five network model to measure the performance based on wrong decision probability [7]. The wrong decision probability was computed based on available bandwidth. Other algorithms that evaluate the performance can be found in [9-21]. In all the algorithms that the authors developed in [7], the mobile nodes were assumed to be in good health. However in practice not all the mobile nodes in the network are cooperative, but can be in other states like failed state, selfish state or malicious state [20, 21]. In this work, a five node network model is developed to consider all the four states, namely, cooperative state, failed state, selfish state and malicious state in calculating the handover probability, unnecessary handover probability, missing handover probability and wrong decision probability based on the criteria of available bandwidth. Next section presents the analytical models for the UHP, HP and WDP. Section III has the general algorithms which are used in calculating the probabilities and section IV presents the simulated results.

Simulation of Vedic Multiplier in DCT Applications

This paper illustrates the simulation of Vedic multiplier in 2D DCT. The input data is first divided into NxN blocks, each block s of 8x8 size and 2-D DCT is applied on each of these 8x8 block and 2-D DCT is applied to reconstruct the image. The proposed 2-D DCT design uses Urdhva Tiryagbhyam a Vedic multiplication sutra and the Simulations with MATLAB prove that the proposed design is compared to that of conventional design. Performing DCT computations using Vedic multiplication sutras gives a significant performance even compared to a DCT using conventional. To illustrate our approach, the sample code implements part of JPEG compression routine, performs forward DCT on 8x8 blocks, quantizes coefficients, and performs inverse DCT.

Meta-Heuristic Approach for Resource Optimization in Mobile Real Time Video Traffic

With the advent of Next generation wireless networks such as 4G, thousands of users will be able to share, create and access live video streaming with different content and characteristics, such as cricket matches and video surveillance using handheld mobile devices. Such services demand new mechanisms for assessing the quality levels of videos. It is imperative for the network operator to exercise stringent control over network parameters and deliver real time video with uncompromising quality in spite of hostile mobile environment. In this paper, Simulated Annealing is used as one of the meta-Heuristic approach for resource optimization in mobile real time video traffic. Other meta-heuristic methods such as Tabu search and genetic algorithms were used in the past. It is found that tabu search gets trapped in local minimum and genetic algorithms are computationally intensive. In this direction, simulated annealing has been found to exhibit effective global minimization with reduced number of iterations. Improvement in performance has been depicted through function plots such as Best function value, Best point, stopping criteria and temperature plot. The algorithm has been tested on standard video quality database available on university at Texas portal. General Terms Meta Heuristic, Optimization.

Texture Feature Extraction through Oblong Aperture and Segmentation using Level Sets

An explorative work on texture feature extraction through oblong aperture for the non random type of texture images is presented in this paper. These features are further useful for segmenting the texture regions using the level set framework. The statistical moment descriptors are obtained within a small aperture and are embedded into a level set frame work for segmentation. The shape of the aperture normally would be square and size selection done optimally. The square shaped aperture sometimes does not yield good feature descriptors particularly when the textured regions are highly structured. The proposed oblong aperture provides an appreciable change in extracted features particularly for structured images in contrast to the square shaped aperture.