Location recognition using detected objects in an image

Several methods have been developed in order to recognize a location from an image. Early methods use appearance based matching, but they usually failed to handle occlusion. Recently, some methods using feature based matching have been developed. They are more robust and faster than the appearance based, but processing time and memory usage are still aspects which can be improved further. Currently, there are a lot of feature-based image matching methods have been developed, such as SIFT, SURF, CenSURE, BRIEF, etc. It has been proved that these methods have been able to reach a high recall rate and yield a satisfying result in object recognition. Using one of the feature-based methods, SURF, we present a fast and efficient location recognition method based on set of objects recognized from an image. We use clustering, greedy N-best paths matching, ratio test, and Hough Transform to improve the object recognition process. Then, properties of detected objects are extracted and matched against rules from training data. The trained rules are built from inter-object properties from the location training images. From our experiment, the result shows a good and promising result. With further development, our method should be able to handle recognition on a large set of location images.