Aligning with Your Own Voice: Self-Corrected Preference Learning for Hallucination Mitigation in LVLMs