A Web Application for Content Based Geographic Image Retrieval

Abstract

In this study, a web application is presented for detecting locations with content-based similarity on high resolution orthoimagery via histogram matching methods. For this application, a web interface utilizing Bing Maps for determining query image was constructed via 'Python Flask Micro Web Framework'. This web interface uses the center location of the Bing Maps extent, and the image closest to the center location is used as a query for histogram based matching with other 29106 images in the collection and presenting the user defined number of similar images with an option to convert KMZ files. In order to determine general performance of the application in dataset, 600 of the images were labeled into 6 classes and with 100 images per class. The application was also implemented by using these labeled images and each of them came into use as a query image. The results were calculated as MAP (Mean Average Precision) of each class containing 100 images for each histogram matching method used in this study.