An Adaptive Approach to Content Based Image Retrieval using Deep Learning

Abstract: Large picture datasets are now possible thanks to developments in image capturing and data storage. In this case, it's essential to create the right information systems to manage these collections effectively. The method in question is known as a content-based picture retrieval system. In essence, these systems attempt to find images that match a user-defined standard or pattern (for example, a form drawing or a mage example). Their objective is to facilitate picture retrieval based on feature vectors that are often stored with content qualities (such as form, colour, and texture). The ability to perform automatic retrieval instead of the extra keyword-based technique, which typically necessitates very hard and time-consuming earlier annotation of database photos, is one of the main benefits of the Content Based Image Retrieval (CBIR) approach. Applications of the CBIR technology include fingerprint identification, biodiversity information systems, digital libraries, crime prevention, medical, and historical research. This project intends to highlight the state of the art of the current research in this field, define the problems and challenges related to the response of CBIR systems, and describe the existing solutions and applications.

Keywords: Deep Learning, Content Based Image Retrieval, Keras, TensorFlow.