Sourav Saha

A Hierarchical Convex Polygonal Decomposition Framework for Automated Shape Retrieval

With the increasing number of images generated every day, textual annotation of images becomes impractical and inefficient. Thus, content-based image retrieval (CBIR) has received considerable interest in recent years. Keeping it as the primary motivational focus, we propose a method which exploits different degrees of convexity of an object’s contour using a multi-level tree structured representation called Hierarchical Convex Polygonal Decomposition (HCPD) tree and the method also uses a special spiral-chain-code to encode the polygonal representation of decomposed shape at every node. The performance of the proposed scheme is reasonably good and comparable with existing state-of-the-art algorithms.

A Computer Vision Framework for Detecting Dominant Points on Contour of Image-Object Through Thick-Edge Polygonal Approximation

This paper presents a computer vision framework for detecting dominant boundary-points on an object’s contour through polygonal approximation of the shape without loss of its significant visual-interpretation. The proposed framework attempts to approximate a polygonal representation of the contour with each polygonal-side having a meaningful thickness to handle noisy curvatures with irregular bumps. The vertices of the polygon are extracted through a novel recursive strategy. The merit of such a scheme depends on how closely it can represent the shape with minimal number of vertices as dominant points without losing its inherent visual characteristics. As per our observation, the proposed framework seems to perform reasonably well in approximating the shape of an object with a small number of dominant points on the contour.

Drunken driving detection and prevention models using Internet of Things

The Internet of Things comprises of a number of uniquely identifiable devices capable of communicating over a network. An individual device in IoT can be used in a wide range of applications. In this paper a model based on IoT is proposed with the aim to safeguard drunk and drowsy drivers especially at night. It also discusses several models which have already been proposed and attempts to assimilate the best ideas which are proposed there. It includes analysis of alcohol concentration, eye-blinking rate and the rate at which the car is made to turn to detect a drunken or drowsy state and hence undertake protective measures. Such measures include speed reduction, triggering an alarm, informing the traffic control, activation of auto-pilot etc.

Flood forecasting using Internet of things and artificial neural networks

Floods are the most common natural disasters, and cause significant damage to life, agriculture and economy. Research has moved on from mathematical modeling or physical parameter based flood forecasting schemes, to methodologies focused around algorithmic approaches. The Internet of Things (IoT) is a field of applied electronics and computer science where a system of devices collects data in real time and transfers it through a Wireless Sensor Network (WSN) to the computing device for analysis. IoT generally combines embedded system hardware techniques along with data science or machine learning models. In this work, an IoT and machine learning based embedded system is proposed to predict the probability of floods in a river basin. The model uses a modified mesh network connection over ZigBee for the WSN to collect data, and a GPRS module to send the data over the internet. The data sets are evaluated using an artificial neural network model. The results of the analysis which are also appended show a considerable improvement over the currently existing methods.

Locating mobile nodes using heuristics with fuzzy logic handoff

In wireless and mobile networks ensuring smooth handoff and tracing exact location of destination node and that too without wasting much time and bandwidth of constrained network is much essential. Several related techniques have been proposed in literature. Each has its own advantages/disadvantages. In this paper, we have proposed, first, a fuzzy logic based handoff management scheme that also handles some basic network management activities like call transferring after allotting a suitable wireless or radio link to the incoming node. It also proposes a location management scheme based on simple heuristics technique like time and probability.

Mobile-autoconf: Mobility management with autoconfiguration in mobile ad-hoc networks

A mobile node needs to be configured with a unique IP address. It has been seen that address autoconfiguration is a suitable technique in MANETs because of high mobility of nodes. Extensive survey of previous autoconfiguration related schemes reveal the need for reducing DoS attack, security breaches and improving DAD, QoS and overall performance. This paper also presents a critical study of mobility management related schemes. In this paper, we propose a scheme Mobile-autoconf that employs ring-based hierarchy for both autoconfiguration and mobility management for larger MANETs. It reduces DAD dependency, scope for DoS attack and temporary address dependency. It reduces cost increases reliability, robustness and location management becomes easier. QoS improvement and recycling of address pool are also aimed at here. The simulation of Mobile-autoconf is also carried out.

A Machine Learning Framework for Recognizing Handwritten Digits Using Convexity-Based Feature Vector Encoding

Handwritten digit recognition has always been an active topic in OCR applications as it stems out of pattern recognition research. In our day-to-day life, character image recognition is required while processing postal mail, bank cheque , handwritten application form, license plate image, and other document images. In recent years, handwritten digit recognition has been playing a key role even for user authentication applications. In this proposed work, we develop a gradient descent ANN model using novel and unique geometric feature extraction technique for handwritten digit recognition system which can be further extended to identify any alphanumeric character images. We have extracted geometric features of handwritten digit based on computational geometric method and applied artificial neural network (ANN) technique for classification of handwritten digits through machine learning approach. The characteristics of extracted feature for a digit class are found to be distinct despite wide variations within the class and thereby lead to reasonably good recognition rate even with small trainee samples.

Unmanned multifunction robot for industrial and military operation over resource constrained networks: An approach

The modern society is now fully dependent upon technology and the technological approach has brought a revolutionary change in each and every field. This paper proposes a multipurpose robot to be used in the battle field. The robot contains Raspberry Pi which acts as a client, packs a video camera for live video streaming, mapping and gripper for disposal of explosives, a Wi-Fi module for controlling the robot remotely from any part of the world by the concept of Wireless Connectivity. Apart from this a control application is developed for the Android based smart phone by which all the functions can be controlled. The unique facility of this system is that the controller can control the device remotely far away from the hazardous battle field. Each operator can conjure up the movement by the live video streaming, enabled by the video camera. By availing this technology necessary precautions and corrections can be laid down by not harming any human lives.

A shape characterization framework for retinal vascular structure analysis

Abnormalities in the vascular structure of a retina, such as abnormal changes in thickness of vessel, tortuosity, or the appearance of retinal lesions, may be associated with the occurrence of retinopathy diseases. An automated structural analysis of changes in vessel morphology may help indicating the clinical signs of retinopathies, describing their early occurrence or severity. We propose a framework consisting of a set of methods for automated characterization of the retinal vessel structure with respect to its morphological properties in two-dimensional fundus images comprising of segmented vascular networks. The methods are validated with the manually annotated retinal fundus images as a ground truth.

Early Scenario Pruning for Efficient Design Space Exploration in Physical Synthesis

Automated design space exploration has become a promising approach for improving VLSI design quality and achieving a balance across a range of design closure objectives. However, it comes with the challenge of high compute resource cost in terms of CPU runtime, disk space, and memory requirements. This paper proposes an automated early scenario pruning (ESP) scheme that predicts the quality of results for synthesis scenarios at the early stages of the synthesis runs. Scenarios deemed non-competitive in terms of objective cost functions can be pruned prior to completion under the proposed framework. This flow reduces the compute resource cost significantly, especially for big and complex design blocks, making design exploration more efficient. The proposed framework is a layer on top of the logic and physical synthesis programs for a high-performance microprocessor design environment. Our experimental data on 14nm processor design blocks suggest over 80% pruning efficiency is possible while claiming approximately 20% savings in CPU time.