Linked Presentation: Towards Targeted Obfuscation of Adversarial Unsafe Images using Reconstruction and Counterfactual Super Region Attribution Explainability