Pre-extraction of Features and Environment Variable-based Database Filtering for Fast Image Matching on Mobile

摘要

Image matching refers to a technique which finds the best matching image from the database to an input image. Currently, feature-based image matching is widely used, and several features have been proposed to reduce the computational cost of image matching. This paper presents a method to further reduce the computational cost of image matching so that image matching can be applicable to mobile services. To this end, instead of storing images to manage the database, we extract features from images in the database beforehand and keep the features. In this way, the computational cost of the actual image matching is reduced since feature extraction is performed on the input image only. Furthermore, we manage the environment variables of each photographing. As most of the current mobile phones have GPS sensor and magnetometer, we utilize these variables to narrow down the matching candidates in the database. The proposed method was applied to outdoor localization, and simulation results show that the proposed method reduced the computational cost for image matching.