Pošalji tekstualnu poruku: Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding