抄録

H-026
Compact Binary Local Descriptor for Object Identification

This paper proposes a binary descriptor for compact representation of feature around local keypoints in an image. The proposed binary descriptor is extracted by binarizing histogram of quantized gradient orientations of a local patch centered at a keypoint. Subset of bits is progressively selected from the binarized histogram forming a descriptor with a size of 32-150 bits. Experimental result using images with planar and 3D objects shows that the proposed descriptor achieves identification accuracy that is comparable with SIFT descriptor, while the descriptor size is less than 1/10.