Abstract

In surveillance, the need for enhanced image representation and rendering from low-resolution videos is paramount for effective analysis and decision-making. This research addresses the limitations of conventional methods in extracting meaningful information from low-resolution footage. The prevalent challenge lies in the compromised clarity and detail inherent in surveillance videos, hindering accurate identification and analysis of critical events. The ubiquity of surveillance cameras has led to an influx of low-resolution videos, limiting the efficacy of traditional image processing techniques. This research aims to bridge this gap by leveraging DenseNet, a densely connected convolutional neural network (CNN) known for its ability to capture intricate features. The DenseNet seeks to enhance the representation and subsequent rendering of images, transcending the constraints imposed by low resolutions. The network ability to capture intricate details will be harnessed to enhance image representation. Subsequent rendering techniques will be employed to reconstruct high-quality images for improved analysis. The results showcase promising advancements in image representation and rendering using DenseNet. The enhanced visual quality of surveillance images allows for more precise identification and analysis of events, demonstrating the potential impact of the proposed methodology on improving surveillance systems.

Authors

D.C. Jullie Josephine, B. Yuvaraj, Sathesh Abraham Leo, S. Thumilvannan
Kings Engineering College, India

Keywords

Surveillance, Low-Resolution Videos, DenseNet, Image Representation, Rendering