Dog nose-print recognition based on the shape and spatial features of scales

This study proposes a novel method for dog identification using nose-print recognition, with applications in controlling stray dogs, locating lost pets, and pet insurance verification. Nose prints offer a unique and safer means of recognition than implanted chips. Accurate positioning of the dog's nose print (DNP) region and comparing its features can enhance recognition accuracy. The two-stage segmentation method effectively segments dog nostrils and nose boundaries in the DNP image. It combines brightness and texture enhancement with the U-Net model to generate a dog nose-print mask (DNPMask) and extracts scale-like features using a genetic algorithm. Experiments comparing one-stage and two-stage segmentation methods demonstrate the latter's superiority, with higher recall, precision, accuracy, and F1 score values. The DNPMask method achieves an average recognition rate of 94.93%, 97.10%, and 97.10% for Top1, Top2, and Top3, respectively, significantly improving dog recognition accuracy. However, its performance may be affected by poor-quality images. Nevertheless, the proposed dog recognition method effectively identifies most dog identities and holds promise for real-life applications, enhancing overall accuracy. The research findings provide a valuable reference for developing dog recognition systems and offer new ideas and directions for related research fields. The improved performance of dog nose-print recognition based on scale-like shape and spatial features (DNPISSFS) supports its superiority over Texture Feature-based Dog Noseprint Image Matching (TFDNPIM), suggesting enhanced dog recognition capabilities with broader implications in animal tracking, pet management, and wildlife conservation. Further analysis and experimentation on a larger dataset will ascertain the method's generalizability and robustness, addressing potential challenges and limitations for future research and practical implementation in various domains. The source code and trained models are publicly available at: https://github.com/Chuen-HorngLin/Texture-Feature-based-Dog-Noseprint-Image-Matching.