Facial expression recognition plays a vital role in various domains, including human-computer interaction, affective computing, and psychological research. However, achieving robust performance across different datasets remains a significant challenge due to variations in image quality, subject demographics, and annotation protocols. This research paper presents a novel technique for facial expression recognition based on Convolutional Neural Networks (CNN) that achieves significant results on cross datasets. By employing transfer learning and database-independent training strategies, the proposed method is trained on the FER-2013 dataset and demonstrates good performance on FERG-DB, CK+, FER-2013, and JAFFE datasets, respectively, of 92.05%, 89.93%, 72.16% and 85.71%. Experimental results show the effectiveness of the technique in achieving database independence and satisfactory performance on diverse datasets.