Support vector machine active learning by Hessian regularization.

•We exploit local structure by using Hessian regularization.•We integrate Hessian SVM into active learning to boost the accuracy in each round.•We conduct extensive experiments including image segmentation and activity recognition.