Motor imagery EEG (MI-EEG) signal identification is used in many brain-computer interface systems. And in most existing brain-computer interface systems, this identification relies on a classification algorithm. However, generally a large amount of subject-specific labeled training data is needed to reliably calibrate the classification algorithm for each new subject. What is more, as MI-EEG signals involve personal privacy information, the security of identification process is very important. But few classification algorithms have paid attention to security. To overcome these challenges, this paper proposes a kernel-ridge-regression-based inductive transfer learning approach (KRR-ITL), which aims to reduce the amount of subject-specific calibration data. It adopts inductive transfer learning to leverage knowledge without directly utilizing samples from an existing subject to improve the calibration performance for a new subject. Experimental results validate that the used approach results in significantly better classification accuracy than many other state-of-the-art approaches in the literature.