Due to complicated structure and interaction of multiple components in rotating machinery, coupling faults have complex dynamic characteristics. Vibration signal analysis has been widely used for fault diagnosis, but it presents difficulty in identifying coupling faults, especially when the coupling faults have similar patterns. On the other hand, infrared image processing can simultaneously diagnose multiple faults with temperature variations, but it is not effective for temperature-insensitive faults. To better utilize multi-modality sensing and to address their limitations, an information fusion method by fusing infrared image and vibration signals is investigated for improved machinery-defect diagnosis in this study. An enhanced non-subsampled contourlet transform (NSCT) method is investigated first for information enhancement and noise reduction of infrared image. Next, feature-extraction strategy on multi-source data is discussed in order to reduce the dimension and enhance fault-representative features for subsequent defect classification. A Dempster-Shafer (D-S) evidence theory-based classifier fusion is then performed to improve the defect diagnosis' accuracy. Experimental studies on a rotor testbed illustrate that the information fusion approach can effectively recognize the coupling faults and improve the accuracy of fault diagnosis in comparison to the methods with single-modality sensing signal.