Affine Collaborative Representation Based Classification for In-Air Handwritten Chinese Character Recognition

This paper presents a study of using affine collaborative representation based classification (ACRC) for recognizing the in-air handwritten Chinese characters, which is collected by Leap Motion, one kind of 3-dimensional sensing device. This classifier uses the form of collaborative representation based classification (CRC), namely it also use the l(2)-norm for sparse coding. But there exists differences between them. Firstly, the atoms of ACRC's dictionary comes from the eigenvectors of each class's covariance matrix, which is easy to obtain in the dictionary learning stage and the size can be much smaller than CRC. Secondly, we compute the coefficients of each class separately on each class's dictionary atoms, instead of computing on the total atoms of the whole classes. For evaluating this classifier's performance, we conduct a series of experiments on the in-air handwritten Chinese character dataset, IAHCC-UCAS2016. And supplementally, we also do the experiments on the online handwritten Chinese character dataset, SCUT-2009. The results are satisfying for the sufficiently high recognition accuracy and the high speed of computation.