13 research outputs found
Natural image matting with non-negative matrix factorization
No full textThis report summaries the work done by the author on his Final Year Project at
Nanyang Technological University (NTU) under Associate Professor Deepu Rajan.
The author was involved in implementing an effective way to realize image matting
using Non – Negative Matrix Factorization.
The report provides an overview of the project. It aims to provide the reader an
insight into the author’s role in the development of the matting algorithm. The report
explains the salient features of matting, Non-Negative Matrix Factorization and how
the entire development process was done. The report summarizes the results found
and compares the obtained results with other known algorithms with the use of
images.
It also highlights the domain knowledge, skills gained to accomplish these tasks and
the challenges faced in the process and how they were overcome. The technical
knowledge gained at NTU, was utilized in various ways to fulfill the requirements of
the project.Bachelor of Engineering (Computer Engineering
Secure Mindmetrics Authentication Mechanism with Log Files
No full textAbstractSecurity is inevitable feature of any web application. Most of the web applications have some kind of design or developmental fault which can be easily exploited by hackers. These faults help them to gain illegal access to the system. Hence majority of attacks are focused on applications thereby stealing user's sensitive data and information.Authentication is an important factor of securityin computing system. Usually login ID's are used for identification and passwords are used for verification. Other mechanisms like secondary or graphical passwords,One time password, Challenge response, Biometric login, Behavioral pattern and Location based authentication can be used.But implementation of these mechanisms requires specialized devices which are not reliable. This paper deals with an enhanced version of Mindmetric authentication mechanism, which grants access only to valid users and helps the admin to determine malicious users with the help of logfiles. It can also be used for preventing users from various attacks. It can augment the current password based system by strengthening the identification process.Mindmetrics system raises security of authentication system over single or double password systems. Even If password is hacked, login attempts by attackers will be blocked by the identification server. It is very simple, scalable and does not require any specialized devices or complex algorithms
Distinguishing amateur and professional photographs
No full textPhotography is the art of capturing and handling images. There are many ways to define the aesthetics in photography. The act of quantifying these aesthetic properties directly to distinguish photographs taken by amateur and professional photographers is almost impossible. This is because there is no general consensus. As such, it is beneficial to develop an algorithm that can differentiate the photographs.
In today’s technological advanced society, there are several researches done by computer scientist and engineers specialised in the field of image processing to learn aesthetic properties of the photographs. The properties are changed into computable image features for classification of photographs.
The project requires the author to understand and implement one of the research papers. The author furthers his reach by deriving new features he discovered upon learning more about photography. This allowed him to improve on the classification accuracy.
In this report, the author explains the various aesthetic appeals of photographs that are used for photograph classifications. The concept of computer vision and image processing to use to extract these aesthetic properties in order to convert into computable data and the concept of machine learning to train a model which is used to differentiate photographs are studied in order to fulfil this project’s requirement.
To determine the feasibility of the improved design, an application is implemented on MATLAB platform. It automatically takes in thousands of already classified photographs taken by professional and amateur photographer as training datasets and another set of randomly chosen picture as testing datasets. The program, once executed, allows the author to differentiate the photographs. The main features, design methodology and test specification of the application are discussed in this report.
Performances analysis of the implemented application is noted. The author also identified further areas that can be enhanced.Bachelor of Engineering (Computer Engineering
Studies on Evaluation of Suitability of Mango Varieties Sindura, Mallika and Totapuri for Processing into Canned Products and Development of Blended Ready to Serve Beverages
No full textThis Dissertation / Report is the outcome of investigation carried out by the creator(s) / author(s) at the department/division of Central Food Technological Research Institute (CFTRI), Mysore mentioned below in this page
Brain-computer interface and visual perception
No full textBrain-Computer Interface (BCI) has been a popular area of research, currently we can find several BCI related product in the market. For example the robotic arms that makes use of BCI.
However, BCI should not be just restricted in controlling robotic limbs and machines, but something that is more between humans, like emotions. The question that the author would like to answer in this project is whether our emotion changes when we look at different images. When we look at a happy photo and sad photo, we felt different emotionally. But is there a difference in our brain signal when we see different things? If yes, is it possible to detect and pick up the differences in term of our brain signal?
In this report, the author will be directing a project in developing a BCI system then find out if there exists a difference in EEG signal and later classify the difference into 1 of the 4 different targeted emotions (Anger, Fear, Happy and Sadness). By using a wireless EEG device, EPOC neuroheadset, to retrieve the EEG signal from a person’s scalp. The author also implements a Matlab script to handle the signal processing tasks in order to obtain the EEG signals.
The author also conducts experiments to test out the BCI system, whether it is capable of classifying the emotion of a person when he/she is looking at images of different objects.Bachelor of Engineering (Computer Engineering
Green Index Quantification of a Unit Manufacturing Process through Simulation Experiments
No full textAbstractIndustries focus on technological and economic growth. Environmental degradation is an adverse outcome of rapid industrial growth. Besides technological and economic objectives; it is equally important to consider environmental impacts as well. This research aims to quantify an eco-efficient manufacturing process, which minimizes resource utilization. Laboratory experiments were conducted to identify significant factors that influence the Green Index of the unit manufacturing process and simulation was used for optimization of these factors settings. Additionally, parameters influencing the work environment adversely were identified and minimized. The results suggested significant improvement in the Green Index of the unit manufacturing proces
Tiny Models are the Computational Saver for Large Models
No full textThis paper introduces TinySaver, an early-exit-like dynamic model compression
approach which employs tiny models to substitute large models adaptively.
Distinct from traditional compression techniques, dynamic methods like
TinySaver can leverage the difficulty differences to allow certain inputs to
complete their inference processes early, thereby conserving computational
resources. Most existing early exit designs are implemented by attaching
additional network branches to the model's backbone. Our study, however,
reveals that completely independent tiny models can replace a substantial
portion of the larger models' job with minimal impact on performance. Employing
them as the first exit can remarkably enhance computational efficiency. By
searching and employing the most appropriate tiny model as the computational
saver for a given large model, the proposed approaches work as a novel and
generic method to model compression. This finding will help the research
community in exploring new compression methods to address the escalating
computational demands posed by rapidly evolving AI models. Our evaluation of
this approach in ImageNet-1k classification demonstrates its potential to
reduce the number of compute operations by up to 90\%, with only negligible
losses in performance, across various modern vision models
Microton irradiation induced tuning of dielectric properties of nano ZnO-natural rubber disks
No full textThe effect of electron beam irradiation of dielectric and conductivity properties of nano ZnO-natural rubber (NR) disks was investigated here. It is revealed that electric properties such as AC conductivity, dielectric constant, and loss tangent of the irradiated samples were improved significantly as compared to the non-irradiated samples, which have been associated with defects in the composites. The total number of dipoles was generated inside the polymer matrix upon irradiation depends on the dislocations formed inside the matrix. From the experiments, we observe that in the amorphous region electron beam irradiation fetches crosslinking and breakdown at the same time. The enhancement of the dielectric and conductivity properties demonstrates that nano ZnO-NR disks will be a promising candidate for the optoelectronic industry. Finally, we also examined the influences of temperature on the electrical conductivity of irradiated samples. 1 2017, Springer-Verlag Berlin Heidelberg.Scopu
Impacts of Urbanization, Aerodynamic Roughness, and Land Surface Processes on the Extreme Heavy Rainfall Over Chennai, India
No full textIn this study, the impacts of urban land surface processes on the extreme heavy rainfall event on 01 December 2015 over Chennai, located in north coastal Tamil Nadu, India are analyzed using convection permitting WRF simulations. A series of numerical experiments are conducted using different land cover data (USGS-1992, NRSC-2004, NRSC-2015), aerodynamic roughness, and land surface models (LSM) to assess their sensitivity on the predicted rainfall. Results suggest that experiments with NRSC-2015 with increased urban extent improved the rainfall prediction in terms of rainfall intensity and its distribution. Though temperatures, sensible heat, and Planetary Boundary Layer height (PBLH) increased due to urbanization in both dry and wet phases, the humidity and Convective Available Potential Energy (CAPE) reduced during the dry phase suggesting thermal convection played a secondary role in rainfall. Considerable increase of surface drag, momentum transport, wind shear and Turbulent Kinetic Energy are found in simulations with updated land use and roughness, which determined the location of the cyclonic circulation, convergence and maximum precipitation. LSM sensitivity experiments indicated that while the five-layer model substantially increased the sensible heat, temperature and PBLH, it reduced the moisture convergence and CAPE relative to Noah and Noah-MP thus resulting in low rainfall. The simulation with Noah-MP enhanced the low-level shear and convergence over other LSMs thus produced a wide spread rainfall along the coast. Our results demonstrated that the momentum transport due to urban drag played a vital role by strengthening the low-level convergence and moist convection, which caused heavy precipitation over Chennai.Authors thank Dr. A.K. Bhaduri, Director of IGCAR, for the encouragement and support. The first author is grateful to HBNI for providing the research fellowship and IGCAR DAE for extending facilities to conduct the study. The updated land use/land cover data over the Indian region are obtained from Bhuvan of ISRO-NRSC (https://bhuvan.nrsc.gov.in/bhuvan_links.php). Authors are thankful to Dr. P.V.N.Rao and Mr. Biswadeep, NRSC-Hyderabad for technical discussions on the NICES landuse/land cover data used in the study. Mrs. Deepu Radhakrishnan is acknowledged for the assistance in GIS aided land cover analysis. Doppler Weather Radar reflectivity images and station rainfall data are obtained from the India Meteorological Department. The GPM IMERG precipitation data is obtained from NASA Global precipitation project (https://gpm.nasa.gov/data/directory)
