Efficient and distinctive binary descriptor for rotated circular image recognition
Dongbo Zhang,Honglei Chen,Feng Yin,Zhiqiang Chen,Hongzhong Tang,Haixia Xu
DOI: https://doi.org/10.1007/s00138-019-01026-8
IF: 2.983
2019-01-01
Machine Vision and Applications
Abstract:Based on pairs of spatial symmetric patches, a novel efficient and distinctive binary descriptor is proposed in this paper for rotated circular image recognition. To achieve rotation invariance during feature computation, a local coordinate system is first found with radial transform technology. On the basis of that, local binary patterns against rotation can be extracted. Meanwhile, the circular image is divided into a set of overlapped annular regions, and pairs of patch description are constructed with histogram within the ring. Finally, the rotation-invariant image description can be obtained by concatenating the ring features from inner to outer. The performance of proposed method is tested with three datasets, i.e., the euro coins, QQ expression and car logo. The test results show that its recognition accuracies reach 100, 100 and 97.07%, respectively, which are superior to the results of traditional methods based on LBP feature. And our method presents competitive performance contrasted to recently proposed LBP invariants, i.e., R_LBP, CS_LBP and PRI_COLBP, and conventional floating and binary descriptors, such as SIFT, SURF, BRISK and FREAK. Moreover, the algorithm is efficient, as single-point feature computation needed only 0.045 ms.