Fairness Explainability using Optimal Transport with Applications in Image Classification