Wavelet Local Binary Pattern (WLBP): A Novel Local Descriptor in Harsh Light Variations
Shekhar Karanwal · 2022 4th International Conference on Inventive Research in Computing Applications (ICIRCA) · 2022
LBP and most of its variants performed well in moderate light changes. But under harsh light changes their performances are not effective. So there is a need of more productive descriptor in harsh light variations. With this note, the proposed work introduces a novel descriptor under harsh light variations so-called Wavelet LBP (WLBP). Initially input image is passed through wavelet decomposition (by using haar wavelet at level 1), which generates four sub-bands. First is approx element and other three are detail elements. Further LBP is deployed to all four sub-bands for feature extraction. It has been noticed during the experiments that histogram of the transformed image doesn't perform impressive in harsh light variations. In contrary, map feature beats the histograms feature comprehensively. Therefore the map LBP features from all four sub-bands are integrated into one framework called as WLBP. FLDA is utilized for the feature compaction and matching is achieved from SVMs. Results are done on YB and EYB datasets, whose images are severely deteriorated by light variations. The accuracy achieved by WLBP is well par from previous histogram based descriptors. WLBP also outperforms all literature methods.