Kernel-Based Multiview Joint Sparse Coding for Image Annotation

Figure 1

Schematic illustration of the proposed KMVJSC approach for automatic image annotation. First, we extract multiple features from all the training images; each type of feature as well as the label matrix is treated as a view. Then, we map the different views into a kernel space, where joint sparse coding and dictionary learning are conducted. To label a test image, we extract its multiple features and reconstruct the test image using the learned atom representation dictionary in the mapped space by joint sparse coding. Finally, the product of atom representation dictionary and the corresponding sparse coefficients gives score for the near neighbors of the test image, and a greedy label transfer scheme is used to get the annotation.