WaveLBP based hierarchical features for image classification

• The new image description is built up using a hierarchy of WaveLBP based features.
• The proposed features are of low dimensionality with dense spatial sampling strategy.
• The image descriptor captures the pixel-level, patch-level and image-level features.

Effective image representation is critical for a variety of visual recognition tasks. In this paper we propose to use hierarchical features for image representation by exploiting the combined strengths of the wavelet transform and LBP (WaveLBP). To be specific, we build up image description under a hierarchical framework based on low-dimensional WaveLBP features with dense spatial sampling, which not only extracts multi-scale oriented features and local image patterns, but also captures multi-level (the pixel-level, patch-level and image-level) features. Experimental results show that the proposed WaveLBP based image description achieves competitive classification accuracies for three different visual recognition tasks.