The detection of emotions in psychological patients creates a pathway that changes the whole picture of mental health treatments to be more efficient through continuous and non-invasive emotional tracking. The work is based on the newest computer vision innovations and combines two distinct yet allied paradigms: one that is interpretable, rule-based heuristic model for real-time classification and another that is a deep-learning pipeline with transfer learning using ResNet to subtly differentiate between facial features. The heuristic model achieves transparent and computationally efficient performance with 98% accuracy on FER-2013 in comparison to the CNN-based model with 89% accuracy, demonstrating resistance to data shortage and overfitting. The proposed framework effectively combines interpretability and scalability to deliver not only precise detection of emotions but also valuable insights that are relevant in clinical practices and could be used for quick therapeutic interventions. The above-stated results actually support the idea of hybrid systems for emotion recognition being a bridge over the gap between complex machine learning methods and their application in the field of psychology.