Sparse Representation On Single Image

Abstract

In recent years, sparse representation of vector signals has been successfully applied in the field of pattern recognition. However, this approach can not be used for single image, as it may require the dictionary to be overcomplete. In addition, the sparse coefficients lack some geometric explanations. This work proposes a novel sparse coding technique on single image. This sparse coding coefficients have explicitly the geometric explanations of images. It depicts the structure information of the image which is robust to variations in illumination, expression, and occlusion. Therefore, the sparse coding coefficients can be used for feature representation of images on small sample case. Experiments on face databases demonstrate the effectiveness of the new sparse coding model.