Akila Karmegam

Akila Karmegam

Auto-generated profile, based on publicly available information

Akila Karmegam

Mechatronics Engineering, Kumaraguru College of Technology, Coimbatore, India

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2015 in Procedia Computer Science
Verified
Journal Article
Open access
KAK. Akila
K. Akila
LJL.S. Jayashree
L.S. Jayashree
VAVasuki A
Vasuki A
Contrast enhancement is an important issue in the field of mammographic image processing. It can be classified into two categories: direct contrast enhancement and in direct contrast enhancement. Indirect contrast enhancement involves in modifying histogram of the image. Histogram equalization (HE) is the simplest indirect contrast enhancement technique which is widely used for contrast enhancement. Many variants of HE are proposed so far. Comparison of these techniques is significantly essential in deciding a suitable algorithm for enhancement and further processing. In this paper we applied few indirect contrast enhancement techniques namely histogram equalization, CLAHE, BBHE, RMSHE, MMBEBHE to preprocess the mammographic images. The performance of the methods is measured using effective measure of enhancement (EME) and peak signal to noise ratio (PSNR)

1 March 2015 in Computer Methods and Programs in Biomedicine
Verified
Journal Article
Closed access
JSJayashree Subramanian
Jayashree Subramanian
AKAkila Karmegam
Akila Karmegam
Elpiniki PapageorgiouElpiniki Papageorgiou
Elpiniki Papageorgiou
NPNIkolaos Papandrianos
NIkolaos Papandrianos
AVA. Vasukie
A. Vasukie

1 November 2015 in Computer Methods and Programs in Biomedicine
Verified
Journal Article
Closed access
Elpiniki PapageorgiouElpiniki Papageorgiou
Elpiniki Papageorgiou
JSJayashree Subramanian
Jayashree Subramanian
AKAkila Karmegam
Akila Karmegam
NPNIkolaos Papandrianos
NIkolaos Papandrianos

30 August 2019 in International Journal of Engineering and Advanced Technology
Verified
Journal Article
Closed access
KAK. Akila
K. Akila
BSB. Sabitha
B. Sabitha
RJR. Jayamurugan
R. Jayamurugan
MTM. Teveshvar
M. Teveshvar
NVN. Vignesh
N. Vignesh
This paper discusses about License plate recognition using digital processing of images, where the image of a vehicle is taken and the number plate is then recognized by various layers of digital image processing. The number plate is then allowed to undergo optical character recognition (OCR), this extracts the data and then compares it with a database containing the details of the vehicle. This allows the user to identify the type of vehicle and the identity of the person who is driving the vehicle. It will denote the user about the registration of the vehicle by comparing it with the database of the registered vehicle in the area. The device will consist of a camera which will take the real time footage of the vehicles and a snap from the video of the vehicle is used to recognize the number plate. The processor will process the images and will display the number of the vehicle and the owner of the vehicle in the display, this is achieved by comparing the number of the vehicle with the previously fed data from the database. This device will provide an efficient way for automating a parking system where there will be no need for a human to interfere with the checking of the vehicle and providing passes for the vehicle.

25 August 2025 in Research Square
Verified
Preprint
Open access
PDF file
KAK. Akila
K. Akila
High-resolution imaging would still play an important role in numerous practical problems like smart surveillance, remote sensing as well as autonomous navigation. Deep super-resolution (SR) models are difficult to deploy to such resource-constrained edge devices because of power constraints, memory capacities, and limited computational capacity. This paper introduces LightSRNet a new light-weight network on the topic of super-resolution networks that optimize efficiency in image reconstruction on embedded processing and IoT systems. The network combines effective building blocks like depthwise separable convolutions, residual feature distillation, and attention mechanisms to make it simpler yet retain the quality of the images. In order to further improve deployment efficiency, the model uses both post-training quantization (PTQ) and quantization-aware training (QAT), including 8-bit and 4-bit precision types. These techniques reduce model size and latency of inference significantly with minor changes in accuracy. It uses a hybrid loss term containing both pixel-based and perceptual ones, the former ensuring structural faithfulness, while the latter structural recovery and texture. The model shows a lot of efficiency in terms of trading performance and computational elegance through its extensive experiments performed on some standard benchmarking like DIV2K and Urban100. Also, the quantized variants of LightSRNet demonstrate competitive values of PSNR and SSIM scales and provide a much lower cost of power and latency using the hardware like the NVIDIA Jetson Nano. The effectiveness of individual elements of the architectures was also confirmed by an ablation. An end-to-end pipeline with ONNX Runtime shows how to work in real-time in a real-world setting. This study proposes a workable and scalable solution to high-quality image enhancement when dealing with edge AI.