Delve into Visual Contrastive Decoding for Hallucination Mitigation of Large Vision-Language Models