Efficient Image Retrieval Using Convolutional Neural Networks and Dimensionality Reduction Techniques

Authors

Krishi patel, Rushi patel, Het patel, Pritamshu singh, Roma mevadawala

Abstract

In the realm of computer vision, efficient image retrieval stands as a cornerstone for various applications, ranging from content-based search engines to medical diagnostics. This project, titled ”Image Retrieval,” endeavour to craft a robust system capable of swiftly locating similar images given an input query. Utilising relevant techniques, including Convolutional Neural Networks (CNNs) for feature extraction and advanced dimensionality reduction techniques such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), alongside a number of machine learning classifiers, the project delves into the intricate realm of image classification and retrieval. The CIFAR-10 dataset serves as the bedrock for training and validation.
The journey commences with the acquisition and preprocessing of the CIFAR-10 dataset, meticulously dissecting images into their constituent red, green, and blue channels. Employing a pre-trained ResNet-50 model, features are extracted with finesse, while LDA deftly navigate the labyrinth of high-dimensional data, distilling crucial insights while mitigating computational overhead.
The journey commences with the acquisition and preprocessing of the CIFAR-10 dataset, meticulously dissecting images into their constituent red, green, and blue channels. Employing a pre-trained ResNet-50 model, features are extracted with finesse, while LDA deftly navigate the labyrinth of high-dimensional data, distilling crucial insights while mitigating computational overhead.
The piece de resistance of the project lies in the realm of image retrieval, where an SVM classifier adorned with an RBF kernel emerges as the virtuoso. Predicting the label of the input image with finesse, akin to a maestro conducting an opus, similar images are summoned forth through the ethereal realm of Nearest Neighbour (NN). These retrieved images, akin to a gallery of masterpieces, serve as a testament to the efficacy and prowess of the retrieval system. In denouement, the project not only illuminates the viability of CNNs and machine learning paradigms in the domain of image retrieval but also lays the cornerstone for further explorations and advancements in the ever-evolving landscape of computer vision and image processing.

Keywords

Convolution Neural Network (CNN), Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), ResNet-50 model, Support Vector Machine (SVM)

Efficient Image Retrieval Using Convolutional Neural Networks and Dimensionality Reduction Techniques. Krishi patel, Rushi patel, Het patel, Pritamshu singh, Roma mevadawala. 2024. IJIRCT, Volume 10, Issue 6. Pages 1-9. https://www.ijirct.org/viewPaper.php?paperId=2410057