<p>Recovering the ground-truth label matrix and generating a prediction function for multi-label learning with missing labels is a challenging task. To improve classification performance, label correlation is often leveraged to recover the missing labels. This paper proposes a new multi-label learning approach with missing labels based on label self-representation. In the proposed approach, the self-representation of labels is utilized to explore correlation among labels and recover the missing labels. Meanwhile, known labels are constrained to remain consistent with their ground-truth values during the recovery process. The resulting optimization problem can be efficiently solved using the ADMM method. Experimental results on seven multi-label datasets demonstrate that our algorithm outperforms other state-of-the-art algorithms in multi-label classification with missing labels.</p>