Published Paper Details:

Locating large picture collections requires the use of content-based image retrieval systems. Convolutional Neural Networks (CNNs) are deep learning algorithms which have upgrade the CBIR with their immense picture feature extraction capabilities. The efficiency of CNNs, ResNet-50, and VGG16 three prominent CNN architectures--for image retrieval is studied in this work. This is investigating a transfer learning-based deep learning framework for CBIR. In this study pre trained models such as CNN, ResNet-50 and visual geometry group (VGG-16) are used to obtain properties from image. These models were trained using large Cifar-10 image datasets. These attributes allow for efficient retrieval based on similarity by capturing the visual properties and semantic content of the images. The study evaluates how effectively these CNN architectures perform in terms of retrieval accuracy. A versatile framework for modifying feature extraction layers is provided by standard CNNs. With its residual connections, ResNet-50 enables deeper structures and better learning. VGG-16's depth can cause overfitting on smaller datasets. The results of this study can help in the development of reliable and effective Content based image retrieval systems. For researchers and practitioners looking to use deep learning for image retrieval applications, the comparison offers valuable insight.