Dog Face Recognition Using Deep Features Embeddings

  • João P. B. Andrade UFC
  • Leonardo F. Costa UFC
  • Lucas S. Fernandes UFC
  • Paulo A. L. Rego UFC
  • José G. R. Maia UFC


Over 470 million dogs are kept as pets around the world. Dogs are owned at an average number of 1.6% per household. The US has the most dog pets, where about 68% of households own at least one pet. Lost and missing dogs are a severe source of suffering and problems for their families. So, this paper addresses the problem of facial dog identification. This technology can benefit many applications, such as handling the missing pet problem, granting pets access to their houses, more intelligent zoonosis control, pet health care, and tracking stray pets. We evaluate a Residual Convolutional Neural Network, specifically ResNet-34, for facial identification in dogs. We tested in DogFaceNet and Flickr-dog datasets with and without two face preprocessing techniques: a central crop and an aligned facial extraction. Experimental results show promising results surpassing the state-of-the-art: 97.6% and 82.8% accuracies for DogFaceNet and Flickr-dog, respectively. Moreover, we also provide recall metrics for the best models.
ANDRADE, João P. B.; COSTA, Leonardo F.; FERNANDES, Lucas S.; REGO, Paulo A. L.; MAIA, José G. R.. Dog Face Recognition Using Deep Features Embeddings. In: BRAZILIAN CONFERENCE ON INTELLIGENT SYSTEMS (BRACIS), 12. , 2023, Belo Horizonte/MG. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2023 . p. 125-139. ISSN 2643-6264.