IMRESTAURANT() MATLAB for Feature-based Restaurant Logo Recognition

This paper discusses the implementation of imrestaurant(), a MATLAB function for feature-based restaurant logo recognition. The algorithm works by first applying SIFT to a logo database and then creating a hierarchical vocabulary tree using K-means clustering. The input to the algorithm is a photograph of an image logo, which is first filtered to reduce noise. Then SIFT is applied to the input image, and the output descriptors are pushed to the vocabulary tree. Each image in the logo database is scored using TF-IDF weights and the ten highest scoring database logos are pairwise matched using the ratio test and RANSAC. The restaurant database image with the largest consensus set is searched for on yelp.com. This algorithm was shown to be robust against input image size, background noise, and shearing. Finally, future work in increasing the accuracy of feature detection as well as optical character recognition is discussed.