Recognizing and detecting faces has been a popular study area in the last few years. Formerly low-level computer vision techniques HAAR Cascade, HOG, etc.—were employed for detection and recognition. However, the accessibility of affordable and effective technology has made it possible to conduct significant studies in this field and a variety of real-world applications from virtual realities and biometric verification to surveillance systems and emotion analysis have drawn a greater interest to this area of study. Convolutional neural networks (CNNs), in especially, are deep learning algorithms that have transformed face detection and identification outcomes. The algorithms created in the fields of artificial intelligence, deep learning, and computer vision in particular are fully utilized by modern CNNs. More advanced techniques have also been developed and readily accessible of large-scale datasets with annotations such as the CelebA dataset and the Labeled Faces in the Wild (LFW) dataset. We present a strategy that leverages the face detectors and extractors of features that currently exist and assesses the approach using measures like precision, recall, and F1-score. This is what we would refer to as a conventional multi-class classification issue. For the purpose of assessing process, we additionally employ an internal dataset, which presents additional difficulties for the detector in terms of face extraction and representing features learning.