Look-Alike Face Recognition Using Deep Learning

The identification of similarities and differentiation between look-alike and non-look-alike features in facial images is a burgeoning research area, holding significant promise for applications in imposter identification. This task is complicated by variations in poses, illumination, and gestures within images of the same individual. Convolutional Neural Networks have emerged as a potent tool for face recognition, demonstrating notable efficiency. This research presents a novel methodology that combines image processing and CNN techniques to predict look-alike facial images with a high degree of accuracy. Achieving an impressive 95% accuracy on a dataset comprising 5000 images, our approach surpasses current cutting-edge methods in the field. In the realm of look-alike face recognition, the IMDB-wiki dataset serves as the foundational dataset. For face recognition, we employed various VGG models, such as VGGFace, VGG16, and VGG19. These models possess the capability to extract diverse facial features. To determine the similarity between two faces, we utilized methods such as cosine similarity, KNN (K-Nearest Neighbors) Euclidean distance, and Manhattan distance. This research contributes to the advancement of effective techniques in the evolving landscape of look-alike face recognition.