Content Based Remote-Sensing Image Retrieval with Bag of Visual Words Representation

Retrieval of images assumes a noteworthy part in various areas including therapeutic determination, biometrics, geological data satellite frameworks, web searching and authentic research etc. At the point, when size of the database increases constantly, the applications including images confront new diculties and signicant issues in indexing, learning and retrieving. We require a productive retrieval system to retrieve images from the vision or audio database. CBIR-Content-based image retrieval is a image retrieval procedure used for retrieving images productively by utilizing low level image features texture, shape and color. In CBIR framework, a query image is described by features within the database. In this report, there are three steps. First, images from dataset are split into training and validation sets. Second, SURF features are extracted of the images and they are represented as bag of visual words using clustering and image indexing. Third, retrieval using cosine similarity. All these steps are carried out on remote rensing images. This technique does not require any relevance feedback for retrieval and it also reduces annotation work with similar results to query.