Landmark Recognition Algorithm Based On Sparse Representation Classification And Extreme Learning Machine

Along with the rapid development of intelligent mobile terminals,landmark recognition attracts increasingly attentions by world-wide researchers in the past several years.Proposing a robust recognition system with a high recognition rate and fast response speed is still necessary.In this paper,we consider the problem of landmark recognition using sparse representation in compressed sensing and ensemble based constrained-optimization extreme learning machine(CO-ELM).To integrate the advantages of these two algorithms,we propose a novel landmark recognition frame combined sparse representation and extreme learning machine.The recently popular spatial pyramid kernel based bag-of-words(SPK-BoW)method is employed for feature extraction and image representation due to its effectiveness in exploiting the spatial layout information for landmark images.Main contributions of the thesis are listed below:(1)Landmark recognition with sparse representation based classification(SRC).After constructing the dictionary with training samples,the recognition of a query landmark image is converted to solving a linear representation problem from an overcomplete equation.This problem can be solved by signal recovery algorithm in compressed sensing.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.Sparse representation based classification is able to achieve high recognition rate in general,but also suffers a low testing speed.(2)Landmark recognition with ensemble based constrained-optimization extreme learning machine(CO-ELM).Constrained-optimization extreme learning machine(CO-ELM)adjusts the objective function by introducing a regulation coefficient as trade-off parameter between norm of output weights and training error.To enhance the recognition performance,the ensemble based CO-ELM is proposed as the classifier.Experiments show the proposed method outperforms the original CO-ELM and support vector machine(SVM)in general.(3)Landmark recognition with extreme learning machine and sparse representation based classification(ELM-SRC).Although sparse representation based classification and extreme learning machine exhibit promising achievements in landmark recognition,respectively,designing a robust recognition system with an accurate recognition rate as well as fast response speed is still challenging.To address these issues,we propose a novel landmark recognition algorithm by exploiting the advantages of sparse representation and extreme learning machine.Experiments show the proposed method achieves a high recognition rate than ELM and a lower response time than SRC.(4)Spatial pyramid kernel based bag-of-words based landmark representation(SPK-BoW).Landmarks are usually located in the central of landmark images while the background information usually distributed in the borders.Conventional BoW algorithms usually ignore the spatial layout feature of landmark images,and therefore can't effectively represent landmark images.To takes full advantage of spatial layout information for landmark images,we adopt SPK-BoW in this article.