A Local Feature Descriptor based on Improved Codebook Model
Qinggang Wu, Zhai Xueming, YUE Baohua · 2020
The commonly used local descriptors often suffer from the sensitivity to image rotation. To overcome such disadvantages, this paper proposes a new local feature descriptor based on an improved codebook of ascending permutation. The proposed descriptor is composed of four steps. Firstly, the original image is uniformly divided into numerous blocks. Then, the pixel values in each block are arranged into column vectors in clockwise. Subsequently, the pixel values in the column vector are sorted in ascending permutation. Finally, such pixel values that arranged in ascending order are used as the local feature descriptor for the given image block. The proposed local features can also be transferred to improve the classical local descriptors. To validate the effectiveness of the proposed descriptor, extensive experiments are conducted on Caltech 101 dataset, and the results demonstrate that the improved model of ascending permutation is more robust to image rotation than original ones.