Content Based Image Retrieval using GPU

Abstract

The measure of visual data accessible in digital format has become exponential as of late. Content based Image Retrieval (CBIR) has turned into a efficient tool for image database management. Today, the requirement for dependable, robotized satellite picture classification and browsing frameworks is higher than at any time in the past. Everyday high volume of remotely sensed information is being gathered and sent by physical satellites for investigation. With the advent of space technology and speed of image acquisition we have proposed a CBIR system which meets the requirement of fast growing data by employing GPU. An interface is provided in which user can give query image and similar images are retrieved from the database.

The work presents a memory and run-time efficient image matching method. The work implements the recently published work [14] and [25] in parallel on NVIDIA GPU card. For feature extraction of remote sensing data parallelism is exploited and substantial amount of speedup in different algorithm is achieved. Furthermore neural network is used for feature selection for optimizing on matching time and retrieval result. To build the semantic gap between user perception and feature representation relevance feedback as user input is used.