MLLM can see? Dynamic Correction Decoding for Hallucination Mitigation