Object Recognition Improvements Obtained Through Saliency-Based Image Enhancement

-, Prachi (2023) Object Recognition Improvements Obtained Through Saliency-Based Image Enhancement. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (4MB) | Preview
Preview	PDF (Configuration manual) Download (1MB) | Preview

Abstract

Novel methods have driven image processing and computer vision advances. Artificial intelligence and machine learning—inextricably linked—are advancing rapidly. This article discusses saliency augmentation, a game-changing approach that highlights an image's most significant aspects. This satisfies the pressing need for higher performance and interpretability while also encouraging accurate model predictions. Based on the principle of saliency mapping, which mimics the way in which human eyes focus their attention and identifies sections of a picture that demand close investigation, saliency augmentation was developed. Instead of just adding more data to an existing dataset, as is done with typical augmentation methods, saliency augmentation seeks to emphasise significant characteristics while simultaneously eliminating unnecessary noise. Extensive trials on the CIFAR-10 dataset utilising a wide range of pre-trained network designs, including VGGNet19, ResNet, MobileNet, EfficientNet, and DenseNet, indicate the efficacy of saliency augmentation in our study. On FashionMNSIT dataset ResNet and VGG19, saliency worked with more accuracy than random erasing. In particular, Saliency-based Gradient Augmentation outperforms both Normal Augmentation and Random Erasing by a wide margin of around 2.6% across all types of models. In this paper, we will cover the groundwork for saliency augmentation, from its methodology to its comparative analyses to the ethical concerns we want to raise and the promising new areas we hope to investigate. By bridging the gap between human visual cognition and computing capabilities, Saliency Augmentation creates a new paradigm in the field of picture data enhancement. The consequences of this go well beyond the realm of computer vision and into many other fields. Saliency Augmentation is a method that helps computers mimic human visual perception.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Kumar, Teerath UNSPECIFIED
Uncontrolled Keywords:	Normal Augmentation; Random Erasing; CIFAR10; MobileNet; DenseNet; VGG19; Saliency Gradient; EfficientNet; ResNet
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Tamara Malone
Date Deposited:	07 Nov 2024 15:56
Last Modified:	07 Nov 2024 15:56
URI:	https://norma.ncirl.ie/id/eprint/7163

Actions (login required)

View Item