An Improved Local Binary Descriptor Based Polar Gridding for Rotation Change

Abstract

Rotation change is widespread in many computer vision applications. Robustness in rotation is an important evaluation for local descriptor. This paper proposes a fast and rotation-robust local binary descriptor using polar location, called rotation-invariant local difference binary descriptor (RLDB). RLDB computes binary test of average intensity and gradient in radial and tangent directions of grid cells partitioned by multiple log-polar grids. The proposed local descriptors need no extra work to deal with rotation specially. Integral image in polar coordinate and lookup table mapping between discretized polar coordinates and image pixel locations are proposed and applied to speed up the computation in the presented descriptors’ construction. The proposed descriptor is compared with some of state-of-the-art local descriptors. The experimental results indicate that the proposed descriptor has a better performance in large rotation change and viewpoint change. The proposed descriptor has comparable distinctive performance and construction speed owing to fast computing in polar coordinate.