HALC is a novel decoding algorithm designed to reduce object hallucinations (OH) in large vision-language models (LVLMs). OH occurs when LVLMs generate text that inaccurately describes objects present in the input image. HALC addresses this issue by integrating a robust auto-focal grounding mechanism to correct hallucinated tokens locally and a specialized beam search algorithm to reduce OH globally while preserving text generation quality. HALC can be easily integrated into any LVLM without additional training. Experimental results show that HALC significantly reduces OH across four benchmarks, outperforming state-of-the-art methods. The algorithm uses adaptive focal-contrast decoding to identify optimal visual contexts for each token, enhancing the accuracy of text generation. HALC also includes a comprehensive evaluation framework that supports comparisons across various OH reduction strategies. The method is effective in reducing OH in tasks such as image captioning, visual-question answering, and more challenging tasks that generalize to novel domains. HALC maintains high-quality text generation while reducing hallucinations, making it a valuable tool for improving the reliability of LVLMs.HALC is a novel decoding algorithm designed to reduce object hallucinations (OH) in large vision-language models (LVLMs). OH occurs when LVLMs generate text that inaccurately describes objects present in the input image. HALC addresses this issue by integrating a robust auto-focal grounding mechanism to correct hallucinated tokens locally and a specialized beam search algorithm to reduce OH globally while preserving text generation quality. HALC can be easily integrated into any LVLM without additional training. Experimental results show that HALC significantly reduces OH across four benchmarks, outperforming state-of-the-art methods. The algorithm uses adaptive focal-contrast decoding to identify optimal visual contexts for each token, enhancing the accuracy of text generation. HALC also includes a comprehensive evaluation framework that supports comparisons across various OH reduction strategies. The method is effective in reducing OH in tasks such as image captioning, visual-question answering, and more challenging tasks that generalize to novel domains. HALC maintains high-quality text generation while reducing hallucinations, making it a valuable tool for improving the reliability of LVLMs.