Although real-world applications of emotion detection using CVDL are manifold, the challenge remains to ensure accuracy and robustness to high levels. This work proposes a CNN model, where the aim is the detection of emotions without any transfer learning or an attention mechanism. Our proposed model used convolutional layers with ReLU activation, batch normalization, max pooling, dropout for regularization, and a softmax output layer to classify seven emotions: happiness, sadness, anger, contempt, disgust, surprise, and fear. We incorporated data normalization, Local Binary Patterns (LBP) for texture-based feature extraction, and Synthetic Minority Oversampling Technique (SMOTE) for handling class imbalance using the CK+ dataset. Data augmentation and the Adam optimizer were applied to improve generalizability and training efficiency. With a result showing 97.42% test accuracy, this model effectively recognizes emotions. It is promising; however, much work remains to improve the performance of this model, especially for those whose facial expressions are less prominent.