Facial micro-expression refers to split-second muscle changes in the face, indicating that a person is either consciously or unconsciously suppressing their true emotions and even mental health. Therefore, micro-expression recognition attracts increasing research efforts in both fields of psychology and computer vision. Existing research on micro-expression recognition has mainly used hand-crafted features, for example, Local Binary Pattern-Three Orthogonal Planes (LBP-TOP), Gabor filter and optical flow. Recently, Deep Convolutional neural systems have demonstrated a high degree effectiveness for difficult face recognition tasks. This paper explores the possible use of deep learning for micro-expression recognition. To develop a reliable deep neural network extensive training sets are required with a huge number of labeled image samples. However, micro-expression recognition is a challenging task due to the repressed facial appearance and short duration, which results in the lack of training data. In this paper, we propose to generate extensive training datasets of synthetic images using data augmentation on CASME and CASME II databases. Then, these datasets are combined to tune a satisfactory CNN-based micro-expression recognizer. Experimental results demonstrate the effectiveness of the proposed CNN approach in image based micro- expression recognition and present comparable results with the best-related works.