Automated landmark-based cat facial analysis and its applications

Area of Science:

• Veterinary Medicine
• Computer Vision
• Animal Behavior

Background:

• Facial landmarks are crucial for human affective computing and are gaining traction in animal studies.
• Geometric morphometrics using facial landmarks objectively assess cat expressions, aiding pain recognition and breed morphology studies.
• Manual landmark annotation is time-consuming, limiting data for machine learning and real-time applications.

Purpose of the Study:

• To develop and evaluate fully automated AI pipelines for cat facial analysis tasks.
• To assess the effectiveness of end-to-end landmark-based systems for practical applications in cats.
• To explore the potential of AI-driven facial analysis for breed and cephalic type recognition, and pain assessment.

Main Methods:

• Developed AI pipelines for automated facial landmark detection in cats.
• Utilized two previously collected datasets of cat faces for analysis.
• Applied the pipelines to three benchmark tasks: automated cat breed recognition, cephalic type recognition, and pain recognition.

Main Results:

• The fully automated pipelines achieved 75% accuracy in cephalic type recognition.
• The pipelines reached 66% accuracy in pain recognition.
• These results indicate the viability of automated landmark-based approaches for cat facial analysis.

Conclusions:

• Landmark-based AI approaches show significant promise for automated pain assessment in cats.
• These methods can facilitate objective morphological explorations in feline species.
• Fully automated systems offer a practical solution for large-scale cat facial analysis and welfare monitoring.