Landmark recognition via sparse representation

Automatically recognizing a place or landmark attracts increasing attentions in recent years due to its wide applications in mobile terminals. In this paper, we consider the problem of landmark recognition using sparse representation in compressed sensing. After constructing the dictionary with training samples, the recognition of a query landmark image is converted to solving a linear representation problem from an over-complete equation. The recent spatial pyramid kernel based bag-of-words (BoW) method is employed for the landmark image representation. Two representative algorithms, namely, the orthogonal matching pursuit (OMP) and the sparse reconstruction by separable approximation (SpaRSA), are adopted to find the sparse representation coefficients. Experimental results conducted on the Nanyang Technological University (NTU) campus landmark database are given to demonstrate the effectiveness of the propose landmark recognition algorithm.