In facial recognition systems, accurate identity verification often depends on the quality and consistency of facial features extracted from images. However, relying solely on neutral facial expressions may not fully capture the range of discriminative information available in human faces. This study looks into whether adding more facial expressions, like Happy, surprise, and disgust, can make face verification systems work better. These improvements have direct relevance for real-world deployments such as border control, online banking, telemedicine, and remote identity verification, where users rarely present a neutral expression during authentication, we use the CK + dataset, which has been reorganized by subject, to test how well fusing embeddings from different expressions works. We compare a baseline method (using a single expression) with a proposed method (averaging embeddings across multiple expressions). We use standard metrics like AUC, EER, and F1 score to test different combinations of emotion sets on three pre-trained convolutional neural network models: MobileNetV2, EfficientNetB0, and ResNet50. When we use more than one expression, especially in complex emotion sets, our results show that verification performance gets better every time. This shows that emotional diversity adds facial features that make subjects more consistent with each other and easier to tell apart. The results point to a promising way to make facial verification systems that are stronger and more aware of expressions.