To improve the efficiency of outsourced storage systems, secure deduplication mechanisms have been introduced. Among these, AES-based encryption is widely used, where packet-level messages are encrypted using a key derived from the message itself. This enables the detection of duplicate data, since identical plaintexts generate identical ciphertexts under the same deterministic setting. AES also supports combined encryption functionality and provides strong security assurances. Cloud computing has emerged as a practical solution for storing and sharing massive volumes of data over networks. Although several approaches have been proposed to ensure data security in distributed storage environments, many of them are closely tied to ciphertext-based techniques. Therefore, this paper presents a cloud-based data collection, sharing, and controlled dissemination scheme designed to preserve multi-owner privacy. In the proposed model, the data owner can securely share sensitive information with a group of authorized users through the cloud.

Keywords : Complex Media, Flexible Access Control, Comfortable Reduplication, Portable Distributed Computing.

Authors : Dereje Weyessa

Title : Scalable Data Deduplication for Privacy-Preserving Media Sharing in Mobile Cloud Environments

Volume/Issue : 2025;2( 2 ( April - June ))

Page No : 30 - 34

The genetic retinal diseases, or Inherited Retinal Diseases (IRDs), make up a large proportion of the childhood vision impairment in the developed world, with the prevalence of one in 3,000 births in the population. Three overlapping challenges hinder timely identification: the incredible genetic heterogeneity of the condition (as many as 280 implicated loci), the invasiveness of traditional electrophysiological evaluations, and the lack of cooperation that is common in very young children at clinical visits. To tackle these challenges, this study proposes a non-invasive Clinical Decision Support System (CDSS) that combines chromatic pupillometry and supervised machine learning to objectively classify the Retinitis Pigmentosa (RP) condition in a paediatric population. A spectrally calibrated red (625 nm) and blue (470 nm) light stimulation was used to elicit monocular pupillary responses in 65 subjects (39 confirmed RP, 26 age-matched controls). Leave-One-Out Cross-Validation (LOOCV) was used to independently optimize the Support Vector Machine (SVM) classifiers, and their outputs were combined using a logical-OR ensemble with the aim of maximizing the sensitivity. A Bidirectional Long Short-Term Memory (BiLSTM) network was also tested to take advantage of the morphology of a temporal signal. The ensemble SVM achieved an accuracy of 84.6%, sensitivity of 93.7% and specificity of 78.6% (AUC-ROC = 0.891). Further, these values were increased to 88.2% accuracy and AUC-ROC = 0.923 by using the BiLSTM model. To the authors' knowledge, it is the first study that combines machine learning with pupillometric time-series data for the field of identification of genetic eye diseases in the paediatric population, thus proof of concept for large-scale screening programmes without contact.

Keywords : Chromatic Pupillometry, SVM, Bidirectional LSTM, Pediatric Screening, Pupillary Light Reflex, PIPR.

Authors : P Viswanatha Reddy , M Anjan Kumar, Praveen Naik , Ramesh Samudrala

Title : Non-Invasive Identification of Inherited Retinal Diseases in Paediatric Patients Through Chromatic Pupillometry and Machine Learning: An Integrated Clinical Decision Support Framework

Volume/Issue : 2025;2( 2 ( April - June ))

Page No : 23 - 29

One of the base functions of the affective computing is facial emotion recognition (FER) which is an ability to understand human psychological state from visual information. In this paper, we propose a hybrid deep learning system that combines a CNN to extract spatial features from a detected face and an LSTM to model temporal emotions, in a real-time face detection system based on YOLOv7. The proposed CNN+LSTM system is trained and evaluated on FER-2013 (35,887 images, 7 classes) and on the test dataset it achieves an accuracy of 89.6%, a precision of 88.2%, a recall of 87.5% and an F1-score of 87.8%, which is a 5.3 percentage-point improvement over the single-frame CNN baseline. YOLOv7 face detector delivers an accuracy of 95.1% on 48 FPS with 93.8% mAP@0.5. Ablation studies validate the distinct contribution of the LSTM temporal module (+5.3%) and of the detection backbone of YOLOv7 (+2.5% with respect to YOLOv5). Facial expression recognition is an integral part of Human-Computer Interaction (HCI) and Affective Computing (AC) which is essential to understand human emotions and respond to them accordingly. In this project, the development and implementation of an automated emotion recognition system using traditional computer vision technique and latest deep learning approach is presented. Firstly, several classical approaches for emotion classification are investigated, including Support Vector Machine (SVM) and hand-crafted feature extractors, such as Local Binary Pattern (LBP) and Histogram of Oriented Gradient (HOG). Despite their effectiveness in some cases, these methods do not scale well to out-of-the-box situations because they cannot apply the input to a variety of facial expressions and environmental conditions. To overcome these drawbacks, the project incorporates cutting-edge deep learning methods, including Convolutional Neural Networks (CNNs) known for their proficiency in extracting hierarchical features from raw image data. Face detection is done in real-time with the help of two object detection models that are widely used in the field of deep learning, namely YOLOv5 and YOLOv7, both of which are considered to be very fast and accurate. Facial regions detected are then fed into CNN models for accurate emotion classification. The system is tested on public datasets and proven to withstand lighting changes, occlusion and cross-cultural differences in expression. Furthermore, privacy and ethical issues are taken into account, highlighting the significance of responsible AI usage. The aim of this research is to contribute to the development of emotion-aware systems that are finding applications in several domains such as healthcare monitoring, online education, smart surveillance, customer service, and interactive entertainment, where understanding the emotions of humans has a positive impact on the user experience and the intelligence of the system.

Keywords : Facial Emotion Recognition, CNN, LSTM, YOLOv7, FER-2013, Affective Computing, Temporal Modelling.

Authors : M Anjan Kumar , P Viswanatha Reddy , T Gopikrishna , MD Akbar

Title : Real-Time Facial Emotion Recognition Using Deep CNN and YOLO-Based Face Detection: A Hybrid CNN–LSTM Framework

Volume/Issue : 2025;2( 2 ( April - June ))

Page No : 14 - 23

Increasing carbon emissions from daily human activities have become a major environmental concern contributing to climate change. Current tools for estimating carbon footprints tend to target small or large user categories and do not provide much assistance to individual users. In this project, you will be implementing the full stack application for measuring, analysing and reducing personal carbon emissions with the help of AI.The system calculates carbon emissions generated from everyday activities such as electricity consumption, transportation, food habits, online usage, product purchases, and waste management. It utilizes intelligent data analysis and recommendation logic to spot emission patterns and make tailored suggestions for the adoption of sustainable practices. An easy-to-use web and mobile application enables users to visualise emission trends, check monthly carbon limits and to see how their emissions are developing over time.EcoTrack's full stack development is paired with the use of applied artificial intelligence techniques to foster environmental awareness and help in informed decision making, empowering users to lower their environmental footprint and help create a more sustainable environment. Full-Stack Architecture combines a responsive front-end with an intelligent back-end, ensuring secure data handling and analytics. The proposed system, through its AI capabilities and data-driven analysis, aims to increase environmental consciousness and facilitate sustainable choices to mitigate carbon emissions. Most people do not have easy-to-use, accurate and effective means to calculate their own carbon footprint, and the amount of anthropogenic greenhouse gas emissions each individual causes by lifestyle choices is a major contributor to climate change. We introduce EcoTrack, a full-stack web application that merges real-time carbon footprint monitoring and personalised carbon footprint reduction suggestions with emission factor-based carbon accounting and leverages the capabilities of Google Gemini 1.5 Pro's Generative AI. Emissions data for four user domains (energy, transportation, diet, and waste) are calculated based on the emission factors exclusively from IPCC AR5/AR6, UK DESNZ 2023, and Poore & Nemecek (Science, 2018). System evaluation showed: Accuracy of emission calculation was 97.3% compared to the manual reference calculation from the IPCC formula; AI recommendation relevance (user survey, n=50) was 88.5%; UI response time was as low as 32.1% of the emission reduction in an average 12-month simulated cohort (n=50); the response time was sub-2 seconds.

Keywords : Carbon Footprint, Greenhouse Gas Emissions, Emission Factor, React/Firebase, Climate Change.

Authors : M Anjan Kumar , P Viswanatha Reddy , T Gopikrishna , MD Akbar

Title : EcoTrack: An AI-Integrated Full-Stack Carbon Footprint Monitoring and Reduction Advisory System for Personal Emission Management

Volume/Issue : 2025;2( 2 ( April - June ))

Page No : 6 - 13

A transient aero-elastic event is encountered by high performance vehicles in general and particularly by Formula class race cars, occurring at millisecond levels, could have a critical impact on the aerodynamic stability and integrity of the vehicle structure. Traditional frame-based camera systems have always been limited because they have a fixed frame rate, causing a time lag, image blur and over-processing the data. This is insufficient for the needs of competitive motorsport where sub-millisecond monitoring is required. In this paper, the authors suggest a Neuromorphic event-camera approach that asynchronously records the intensity change of each individual pixel with microsecond temporal resolution, without any inter-frame redundancy whatsoever. The proposed architecture is a combination of dynamic vision sensor (DVS) event streams and hierarchical spiking neural network (SNN) processing pipeline deployed in an edge compute node, for a real-time aero-elastic instability prediction. Experimental testing at various vehicle speeds of 80–340 km/h in this vehicle demonstrates the capability for the proposed system to obtain 97% diagnostic accuracy and process the data in as little as four milliseconds end to end – a factor of 91% less than conventional frame-camera baselines. At the same time, power consumption is decreased by 79% compared with CNN approaches, which makes the architecture suitable to the edge use case for power-constrained deployment.

Keywords : Neuromorphic Computing, Event Camera, SNN, Aero-Elastic Monitoring, Edge AI, Dynamic Vision Sensor.

Authors : Lava Kumar Vandrangi

Title : Neuromorphic Event-Camera Systems for Ultra-Low Latency Aero-Elastic Monitoring in High-Speed Vehicles

Volume/Issue : 2025;2( 2 ( April - June ))

Page No : 1 - 5