Local Feature Descriptor based on High-Order Co-Occurrence of Gradient Orientation Histograms

Ryosuke Hayashi, Shuichi Enokida, and Toshiaki Ejima


Object Detection, Feature Descriptor, Co-occurrence Feature, Gradient Orientation Histogram


Object detection algorithms have attracted attention for use in image processing. Object detection algorithms are generally divided into two parts. The first part is to calculate the amount of a feature of the input image. The second part is recognition based on the image feature. In the present paper, a new feature descriptor is proposed to calculate the image feature. The proposed descriptor uses a gradient orientation histogram in a local area in a similar manner to other recent methods. In order to increase the object recognition/classification ability, the proposed descriptor calculates the high-order co-occurrence of the histogram elements. In addition, in the present paper, the performances of a number of operators, such as the summation operator, are compared. The proposed descriptor is robust with respect to occlusion/variance problems because the descriptor is able to obtain the informative feature despite observing a localized region of an object image. In the present paper, the performance of the proposed descriptor is demonstrated experimentally.

Important Links:

Go Back