Abstract:  An image classification algorithm based on sparse coding and multiple kernel learning (MKL) was proposed.  First, D-SIFT (Dense Scale-Invariant Feature Transform) and D-SURF (Dense Speeded Up Robust Feature) are extracted from images. Then, sparse coding method is adopted to represent an image as a vector and max pooling method is also utilized for both features. Finally, an improved MKL is used to classify those vectors. Appropriate kernel combinations are selected for each feature and the final result is the fusion of both. The experiments demonstrate that the algorithm remarkably improves the classification accuracy.  

Key words: sparse coding, multiple kernel learning(MKL), feature fusion