Euler Label Consistent K-SVD for image classification and action recognition

Motivated by the fact that kernel trick can capture the nonlinear similarity of features, which may help improve the separability and enlarge the margin between nearby data points, we present an effective kernel dictionary learning approach, namely Euler Label Consistent K-SVD (ELC-KSVD), for sparse coding and image recognition. ELC-KSVD first maps the images into the complex space by Euler representation, which has a negligible effect for outliers and illumination, and then learns a discriminative dictionary in Euler space. Different from the most existing kernel dictionary learning approaches, which maps data into a hidden high-dimensional space, Euler representation not only is explicit but also does not increase the dimensionality of image space in our ELC-KSVD. This makes ELC-KSVD algorithm efficient and easy to be realized in real applications. Furthermore, an iterative method is provided to solve ELC-KSVD. This iteration algorithm is fast and has good convergence. Extensive experimental results illustrate that ELC-KSVD outperforms some representative methods and achieves impressive performance for image classification and action recognition.