Separated Fan-Beam Projection with Gaussian Convolution for Invariant and Robust Butterfly Image Retrieval

Xin Chen, Bin Wang, Yongsheng Gao

Abstract: Highlights•A new concept, named separated fan-beam projection, is introduced for the first time into image recognition domain.•Separated fan-beam projection with Gaussian convolution is designed to construct multiscale invariant features, patch-projection angles (PPA) and texture-projection angles (TPA).•PPA and TPA features are then treated as heterogeneous co-occurrence patterns and fused by 2D histogram as final feature representation.•Extensive experiments with a comprehensive evaluation are conducted to validate the superior performance of the proposed method over the state-of-the-arts for butterfly image retrieval.