Embedding capacity, image quality and embedding security are three of the most important performance indicators for image-based reversible data hiding. Over the past few years, extensive studies have been carried out in improving embedding capacity and image quality. However, little progress is made in enhancing embedding security, which represents the robustness to detection from modern steganalyzers. In this paper, the embedding security against modern steganalyzers is improved by exploiting multiple histograms modification (MHM) to hide secret bits into the complex image regions. Firstly, embedding cost is employed to measure the statistical distortion caused by data embedding, and the data are embedded into pixels with low embedding cost. Secondly, the overall embedding distortion for a given payload is derived as the sum of embedding cost of the modified pixels. Finally, the overall embedding distortion is minimized using MHM. Experimental results show that, the proposed method can significantly improve embedding security, well preserve image structure similarity, and provide a considerable amount of payload.