Boosted Cross-Domain Dictionary Learning for Visual Categorization

Abstract

In an extension of the AdaBoost and transfer AdaBoost algorithms, a boosted cross-domain categorization framework works with a learned domain-adaptive dictionary pair and boosted classifiers so that both the auxiliary domain data representations and their distributions are optimized to match the target domain. By iteratively updating weak classifiers, the categorization system allocates more credits to similar' auxiliary domain samples, while abandoning dissimilar auxiliary domain samples. The authors evaluated the proposed approach using multiple transfer learning scenarios, including image classification, human action recognition, and 3D object recognition. The proposed method consistently outperformed the state-of-the-art methods in all the evaluated scenarios.