Improved structured sparse PCA for cluster-based co-saliency detection
Shen Ningmin, Jing Li · 2015
A novel structured sparse PCA with feature selection refinement for cluster-based co-saliency detection method is proposed. An image is divided into blocks firstly, then the low-level features of RGB and LAB space are extracted, sparse PCA is applied to feature reduction, then the feature selection for loadings refinement is used to improve the results of sparse PCA. Finally, the reduced features is fed to the cluster-based co-saliency detection. This method is unsupervised to implement and efficient to run. Experimental results show our method stably achieves similar performance compared with the cluster-based co-saliency methods with high runtime efficiency. Moreover, our method is shown to be advantageous in co-saliency detection with the noisy image.