Mitigating Hallucinations in Large Vision-Language Models with Internal Fact-based Contrastive Decoding