Histogram intersection kernel for image classification

Evaluation of gender classification methods on thermal and near-infrared face images

“…However, from the results reported in [4] and in subsequent works, it appears that histogram intersection is an effective kernel representation scheme making it a good candidate for building gender classification systems. It has also been shown that histogram intersection has the required mathematical properties for it to be a suitable kernel function for SVMs [4] compared to the Gaussian RBF kernel. Besides, it requires fewer parameters to tune in order to achieve the optimal performance.…”

Mid-level feature set for specific event and anomaly detection in crowded scenes

“…Furthermore, only the best N points (80 in our experiments), those exhibiting the highest corner strengths, are considered for the subsequent tracking process. Next, HOG [13] is used to describe the local area around each detected point, and the Histogram Intersection (HI) [14] is used for feature comparison. The output of this stage is a set of pairs (p,u), where p represents a key-point and u its associated motion vector, from which we build the proposed set of features.…”

Robust Kernel Representation With Statistical Local Features for Face Recognition

“…[34] or Chi-square kernel could be adopted in the l 2 -norm distance based classifiers such as SRC and CRC. However, directly applying these kernels to SLF based representation may not be robust to facial occlusions.…”