Abstract
|
This paper presents a new coding pattern based on conventional LBP in hexagonal lattice using cyclic sets referred as compressed hexagonal local binary pattern (CHLBP). For this task, first, conditions of having a consecutive sequence of 1’s in patterns are investigated. Then, conventional LBP patterns are divided into three sets and then each set is mapped into a new set using cyclic set theorem. Finally, feature vector is defined based on new patterns. CHLBP has been evaluated on four datasets and compared with several state-of-the-art methods with respect to two criteria precision and recall. The best performance of CHLBP is achieved in ORL dataset with the precision of 89.51%. Also, CHLBP is by far the best method in KDEF dataset which improves the precision of the best previous method by 7%. Finally, from runtime and feature vector length point of view, CHLBP is the fastest method in all datasets.
|