GEOMIR2K9 - A Similar Scene Finder

The main goal of the GEOMIR2K9 project is to create a software program that is able to find similar scenic images clustered by geographical location and sorted by similarity based only on their visual content. The user should be able to input a query image, based on this given query image the program should find relevant visual content and present this to the user in a meaningful way. Technically the goal for the GEOMIR2K9 project is twofold. The first of these two goals is to create a basic low level visual information retrieval system. This includes feature extraction, post processing of the feature data and classification/ clustering based on similarity with a strong focus on scenic images. The second goal of this project is to provide the user with a novel and suitable interface and visualization method so that the user may interact with the retrieved images in a natural and meaningful way. 1. Introduction Our goal is to find similar scenic images clustered by geographical location and sorted by similarity based only on their visual content. The user should be able to input a query image, based on this given query image the program should find relevant visual content and present this to the user in a meaningful way. Technically the goal for the GEOMIR2K9 project is twofold. The first of these two goals is to create a basic low level visual information retrieval system. This includes feature extraction, post processing of the feature data and classification/ clustering based on similarity with a strong focus on scenic images. The second goal of this project is to provide the user with a novel and suitable interface and visualization method so that the user may interact with the retrieved images in a natural and meaningful way. There has been significant prior research [1‐13] which has explored a wide variety of approaches from Markov random walks [1] to fusion of features [13]. 2. Method and Ideas To extract low level features that can give meaningful information for a similarity measure aimed at retrieving scenic (and outside) images taken at a multitude of locations from all over the world the following steps were taken. The images are first resized to a standard size to give representative counts when generating the histograms (explained below). Pictures that were made on the side were detected and adjusted accordingly resulting in pictures of size 640x480 and 480x640 …