Hair Style Recommendations using CNN and AR

This research begins with the collecting of a varied dataset of annotated facial photos to ensure accurate representation across populations. The photographs will be pre-processed to standardize their size, illumination, and orientation. On this dataset, we will fine-tune a pre-trained CNN, such as ResNet50, to identify various face forms before evaluating accuracy against a separate test set. Furthermore, we will create a dataset of facial photos combined with various haircuts classified by face shape, and use the CNN to extract relevant features and embeddings. A matching algorithm will match user facial features to hairdo embeddings, ranking recommendations according to similarity metrics. The AR component will leverage ARKit or ARCore for real-time face tracking, allowing users to realistically test on different haircuts overlayed on their live camera feeds. The final solution will be available on mobile and online platforms, with a user-friendly design and options for saving preferred haircuts and providing feedback. This project tackles the demand for a more engaging and individualized hairstyle picking experience, utilizing advanced AI and AR technologies to improve customer happiness and interactivity.