The SIFT image feature reduction method using the Histogram Intersection Kernel

This paper shows that the SIFT image feature descriptors can reduce computational power and memory by introducing the histogram intersection kernel. Many different dimensionality reduction or compression methods for image feature points in the SIFT descriptors allows the computational processes to be more efficient and beneficial for data reduction. Also, this approach is applicable when combined with the feature dimension reduction method. An experimental result shows that a reduction in the feature data size can occur without sacrificing recognition accuracy.