Project Showcase
FISAT
CSE Department
Visual Question Answering
Aathira Satheesh, Anagha K , Fathima Salim , Jerin Jayaraj
Given an image and a natural language question about the image, the task is to provide an accurate natural language answer. Mirroring real-world scenarios, such as helping the visually impaired, both the questions and answers are open-ended. Visual questions selectively target different areas of an image, including background details and underlying context. As a result, a system that succeeds at VQA typically needs a more detailed understanding of the image and complex reasoning than a system producing generic image captions. Moreover, VQA is amenable to automatic evaluation, since many open-ended answers contain only a few words or a closed set of answers that can be provided in a multiple-choice format.
Smart Department
Ajish Antony , Amal Alby , Gokul Babu, Jill Antony
Across the world there are many fields where a lots of computers and devices are used and no proper mechanisms are there to monitor and control these systems. Smart department system is developed on the basis of this to monitor and control systems which are connected together to a server via network. Smart Department System is a powerful system, which provides the users with a rich and responsive user interface and to gather information regarding resources of system within a department. We also provide a proper database system for storing these information. In addition, we provide overall monitoring of system that are connected to a server in department network. Admins are given a privilege to add or delete systems. The system allows different types of user to gather information regarding client systems within a department. Also the admin users are provided with a provision to shutdown a system by monitoring system usage of processor, memory, etc. To be precise the Smart department system can be said to be as an efficient system which can be implemented to monitor and control systems which are connected together.
Dependence Models for Searching Text in Document Images
Ahameed Razal Kapil ,Ameer Iqbal, Anush S, Arjun Shaji
IImage descriptions needs a detailed understanding of the various elements of the image. The elements include the objects/person present in the image , the background or the setting of the environment in which the image is based , and the relationship of the objects and all the entities of the image with among themselves and the environmental setup in which they exist. Language or any form of communication can be used to describe the significant amount of information present around us in the world. Similarly, the language can be used to provide usable and important information from the scenes depicted in the images. This leads to a better understanding of the scene by generating captions out of images and using the captions to thoroughly understand the information from the images. A large amount of information is stored in an image. Everyday huge image data is generated on social media and observatories. Deep learning can be used to automatically annotate these images, thus replacing the manual annotations done. This will greatly reduce the human error as well as the efforts by removing the need for human intervention The generation of descriptions from images has various practical benefits, ranging from aiding the visually impaired, to enabling the automatic, cost-saving labelling of the millions of images uploaded to the Internet every day, recommendations in editing applications, beneficial in virtual assistants, for indexing of images, for visually challenged people, for social media, and several other natural language processing applications.
Improving Home Automation Security;Integrating Device Fingerprinting
Abin Peter , Alen Jose , Daniel Paul , Harikrishnan Menon
Home Automation System (HAS) has gained popularity due to communication technology advancements. Smart home is one of the Internet of Things (IoT) applications that facilitates the control of home appliances over the Internet using automation system.Portable devices are tempting targets for an attacker, as cellphones and tablets connect to different networks as they change location. If the user is unaware of basic security, their smartphone can easily connect to a fraudulent access point. Even some tech-savvy people are concerned with the physical security of their doors and windows, but can be careless when it comes to their digital devices.This paper explains the importance of accessing modern smart homes over the Internet, and highlights various security issues associated with it. This paper explains the evolution of device fingerprinting concept over time, and discusses various pitfalls in existing device fingerprinting approaches. In this paper, we propose a two-stage verification process for smart homes, using device fingerprints and login credentials, which verifies the user device as well as the user accessing the home over the Internet. Unlike any other previous approaches, our Device Fingerprinting algorithm considers a device’s geographical location while computing its fingerprint. In our device identification experiment, we were able to successfully identify most of the devices that visited our Web page using JavaScript, Flash, and Geo-location.
Detection of Nutrient Deficiency in Plants
Aiswarya K A ,Don Paul Jolly, Elizabeth Josy , Jishnu S Prasad
Nutrient deficiency is a common condition that can spread and affect plants,if it is not handled quickly. Normally, deficiency surveillance is carried out manually which needs more effort especially for large area. Nutrient deficiencies cause symptoms such as leaf yellowing or browning, sometimes in distinctive patterns.This project performs a deep convolutional neural network to diagnose deficiency based on image of plants. Inception-Resnet architecture with transfer learning from model that is previously trained using the dataset is applied in the experiments. The dataset collected includes healthy and nutrient deficient images of rose,mango,guava and corn.We have included sulphur deficiency in rose,zinc deficiency in mango,potassium and phosphorous deficiency in guava and also magnesium deficiency in maize.The aim of this project is to predict the nutrient deficiency accurately in order to increase crop production and prevent the emergence of diseases related to nutrient deficiency.We found out the nutritional deficiency and also provide a suitable solution to treat this problem.
Texture Synthesis Over Arbitrary Manifold Surfaces Using Masking
Amal Jacob Regee, Anila Agnes Sheo, Ashwini Harikumar, Hrishikesh Kamalakshan
In computer science, digital image processing is the use of a digital computer to process digital images through an algorithm. Neural style transfer is an optimization technique used to take two images- a content image and a style reference image (such as an artwork by a famous painter). This technique blends the two images together so the output image looks like the content image, but “painted” in the style of the style reference image. The implementation is done by optimizing the output image to match the content statistics of the content image and the style statistics of the style reference image. These statistics are extracted from the images using a convolutional network.
Novel Strategy for Road Lane Detection and Tracking Based On a Vehicle’s Forward Monocular Camera
Annlyn Anto, Jiyadh K Salim , Geo Paul
Modern road vehicles are employing features of driver assistance systems (DAS) to improve drivability performance, comfort, and safety. In the future perspective, the advances in this field will lead these systems to the level of autonomous and cooperative driving, based on sensors networks and sensor fusion. This paper aims to present the readers a novel strategy for lane detection and tracking, which fits as a functional requirement to deploy DAS features like Lane Departure Warning and Lane Keeping Assist. To achieve the presented results, the digital image processing was divided into three levels. At the low-level, the input image dimensionality is reduced from three to one layer, the sharpness is improved, and region of interest is defined based on the minimum safe distance from the vehicle ahead. The feature extractor for lane edges detection design is part of the mid-level processing. The lane tracking strategy development is discussed in the high-level stage; Hough Transform and a shape-preserving spline interpolation are used to achieve a smooth lane fitting. The experimental outcomes were qualitatively and quantitatively evaluated using a ground truth comparison. The strategy shows good accuracy levels, including scenarios with shadows, curves, and road slope.
Algorithmic Analysis On Static and Dynamic Rule Generators
Anagha Zachariah C, Anjo Martin, Angel Louis, Gautam Balamurali
In the last two decades, thanks to dramatic advances in artificial intelligence, computers have approached or reached world-champion levels in a wide variety of strategic games. The work presented here is an attempt to apply various Artificial Intelligence algorithms in games with static and dynamic rules. The first contribution is the development of Rumble Board. Rumble board is a dynamic strategy game in which, the rules are generated in each game. Rumble board is a two-player strategy board game that was designed to be playable with a standard chess set. In Rumble board each player has to conquer opponents territory by killing opponents chief. Each territory is protected by the Marios in which some of them are gifted. The gifted Marios can be identified only after the war starts. So the rules for the movement of Marios are different in each game. This non deterministic nature of the game is analysed by various algorithms. The second contribution is the development of checkers. Checkers is a strategy board game for two players which involve diagonal moves of uniform game pieces and mandatory captures by jumping over opponent pieces .The rules are static and the game is deterministic in nature. The work presented here analyses the performance of minimax, minmax with alpha-beta pruning and reinforcement learning algorithm in Rumble board and checkers.
A Shoulder Surfing Resistant Graphical Authentication System
Aparna T.M, Ashika Varghese, Babitha Raju, Deepthy Paulose
Authentication based on passwords is used largely in applications for computer security and privacy. However, human actions such as choosing bad passwords and inputting passwords in an insecure way are regarded as "the weakest link" in the authentication chain. Rather than arbitrary alphanumeric strings, users tend to choose passwords either short or meaningful for easy memorization. With web applications and mobile apps piling up, people can access these applications anytime and anywhere with various devices. This evolution brings great convenience but also increases the probability of exposing passwords to shoulder surfing attacks. Attackers can observe directly or use external recording devices to collect users credentials. To overcome this problem, we proposed a novel authentication system PassMatrix, based on graphical passwords to resist shoulder surfng attacks. With a one-time valid login indicator and circulative horizontal and vertical bars covering the entire scope of pass-images, PassMatrix offers no hint for attackers to figure out or narrow down the password even they conduct multiple camera-based attacks. We also implemented a PassMatrix prototype on Android and carried out real user experiments to evaluate its memorability and usability. From the experimental result, the proposed system achieves better resistance to shoulder surfng attacks while maintaining usability.
Navigation Assistance For The Blind
Becky Ebby George, Aiswariya Gopakumar, Alice Jobi, Diya Ann V Alias
In India, there are forty million people who are blind. In their day-to-day life, they face a lot of problems like navigation, social challenges and they strive hard to become independent among themselves. To solve this problem we came up with this project. Navigation assistance for blind refers to the systems that can assist or guide people with vision loss, ranging from partially sighted to blind. The obstacle detection, object recognition, emergency message, and navigation always remains a challenging fact. To address these issues, we used video streaming in real-time and filters the unwanted area and detect and recognize the predefined objects and notifies the user. If the user gets into trouble or if need any help the user can give an emergency mail to the guardian. For navigation, it identifies the intersecting line and the direction so that it can lead the user to the destination safely.
Breast Cancer Classification using Machine learning
Gayathry S, Hridya Soman, Fathima Shirin K A
The Invasive Ductal Carcinoma (IDC) is the most common type of breast cancer. We used the histology images of the cancerous cells inorder to classify the cancer into Benign (non-cancerous) and Malignant(cancerous). The classification algorithm used is Convolutional Neural Network(CNN). The neural network is designed using the concepts of Keras and Deep Learning. Our model was able to classify the histology images successfully into positive and negative ones with an accuracy of 85%. We also used OpenCV for the visualizationand processing of images.
Brain Tumor Detection using Deep Neural Network
Abhirami G, Aiswarya Santosh , Amita Ajay , Gopika Raj K
Brain Tumors like Glioblastomas can appear anywhere in the brain and have almost any kind of shape, size and contrast. Due to this feature, there is a need for the exploration of a machine learning solution that exploits a flexible, high capacity DNN while being extremely efficient. We model a novel CNN architecture which differs from those traditionally used in computer vision. The CNN exploits both local features and global features. In difference to most traditional approaches of CNN, the network used here includes a final layer that is a convolutional implementation of a fully connected layer. There is also a 2-phase training procedure that allows us to tackle difficulties with regard to the imbalance of tumor labels.In addition to these, there is a final cascade architecture in which the output of a basic CNN is treated as an additional source of information for a subsequent CNN.
Metaheuristic Algorithms for Global Optimisation
Abijith Pradeep, Aswin Kumar KS, Dileep Vijayakumar
Meta-heuristics are a set of intelligent strategies to enhance the efficiency of heuristic procedure by finding the global or near-optimal solutions at a reasonable computational cost. We are introducing two new innovative meta-heuristic optimisation algorithms - Global Tree optimisation and Iravatham optimisation. Global Tree optimisation is a general purpose optimization algorithm while Iravatham optimisation is focused mainly on scheduling applications. We analyzed the performance of the algorithms in a simulation of cloud computing infrastructures and services using CloudSim and proved that the introduced algorithms are very efficient
Inverse Kinematics using Machine Learning
Bryan Davis
This project aims to create a solution to the problem of Inverse Kinematics using various machine learning techniques. The 2 Major approaches pursued within this paper are The Feed Forward Network Solution (FK) and The Recurrent Neural Network Solution (RNN). The RNN solution in particular is more of a cutting edge proof of concept and its efficacy remains to be seen. The FK model on the other hand is a more tried and tested model and can be shown to work.
An Approach For Autonomous Waste Management
Diya Merin Sunny
Modern cities are growing ecosystems that face new challenges due to the increasing population demands. One of the many problems they face nowadays is waste management, which has become a pressing issue requiring new solutions. Swarm robotics systems have been attracting an increasing amount of attention in the past years and they are expected to become one of the main driving factors for innovation in the field of robotics. The research presented in this paper explores the feasibility of a swarm robotics system in an urban environment. By using bio-inspired foraging methods such as multi-place foraging and stigmergy-based navigation, a swarm of robots is able to improve the efficiency and autonomy of the urban waste management system in a realistic scenario. To achieve this, a diverse set of simulation experiments was conducted using real-world GIS data and implementing different garbage collection scenarios driven by robot swarms. Results presented in this research show that the proposed system outperforms current approaches.
Audio Recognition Using Deep Learning
Arun Raju, Dony Doshy, James Baby, Jeffrey Viju
Audio signal processing is the processing of an audio signal inorder to achieve a certain goal. Gender recognition is one such goal, where when given an audio the system classifies the given input as either a male or female. A gender recognition system can be a base for various other systems like a gender based personal assistant where the assistant can give salutations as well provide any assistance based on the gender. This also can be a base for speaker identification. A gender recognition system when combined with image recognition can be then used to provide tags on an image or a video. This system can be used for security purposes as well. This system is an area with wide range of applications.
Underwater Image Enhancement
Alaikha P V, Anamika L S, Anjana A Lakshmi, Athira C H
Underwater images have low contrast, blurriness and colour cast due to various effects like absorption, scattering and refraction. These results in underwater images being less clear and not trustworthy for activities like underwater image exploration, marine research, automated underwater vehicles and so on.There are numerous methods available for the enhancement of underwater images, but most of them sacrifice one factor for meeting others. In our project, we propose Underwater GAN (UWGAN) and ResNet based method, to enhance the underwater image.Here Underwater Resnet (Uresnet) is used.UWGAN is used to generate realistic underwater images from in-air images.ResNet is a residual learning model for underwater image enhancement task, the loss function is also improved.