Illumination-Insensitive Line Binary Descriptor Based on Hierarchical Band Difference

Xinyu Lin, Yingjie Zhou, Yipeng Liu, Ce Zhu

Abstract: Similar to points, lines are also the basis for many vision applications. Line segments provide more structured information and are more robust than point features, especially for those scenarios with low textures. There are increasing works to tackle the line matching problem, but few of them are designed to match lines under drastic illumination changes, which is critical for long-term vision tasks. In this brief, an illumination-insensitive line binary (IILB) descriptor is proposed to tackle the above issue based on the band differences among multi-spatial granularities of the line region of support. The proposed IILB descriptor has a low computation cost due to the usage of integral images, which could satisfy the requirements of many realistic platforms with limited computation and storage abilities. Numerical experiments in natural and synthetic datasets show that the proposed IILB descriptor has remarkable performance. The code of the IILB descriptor is publicly available.