IMAGE CLASSIFICATION BASED ON ORDINAL REGRESSION AND DICTIONARY LEARNING

To overcome the problem of class order and data redundancy in image classification, this paper proposes a novel implicit constraints support vector ordinal regression method based on dictionary learning (IMCDL). This method aimed to seek a series of parallel hyperplanes to separate the ordered classes, such that the ordered information could be considered to improve the learning model. Furthermore, IMCDL introduced the dictionary learning into the ordinal regression which made the transformed data more discriminative. The experimental results on real-world image datasets show that IMCDL obtains better performance than the existing methods.