<p>Most existing database watermarking algorithms cause permanent distortion to original carrier data, and it is difficult to achieve a balance between watermark embedding capacity and data distortion. For this reason, reversible data watermarking technology has emerged, which can not only extract watermark information but also restore the original data without any distortion. This paper proposes a robust and reversible database watermarking method with low distortion and high capacity, by using Square prediction, Dual-layer embedding, and Histogram Shifting Watermarking (SDHSW) strategies. First and foremost, it utilizes square prediction to divide database tuples into two groups and constructs prediction-error histograms in each group respectively. Next, for each prediction-error histogram, a method with single-bin embedding 2-bits information is employed to expand the watermarking capacity. Subsequently, a scrambling algorithm is used to randomly distribute the attribute values of database tuples, making the distribution of prediction-error histograms more discrete. Lastly, the selection rules for watermark embedding carriers are optimized, effectively eliminating redundant distortion caused by histogram shifting. The experimental results demonstrate that the proposed method exhibits minimal data distortion and a high watermark embedding capacity. Compared with advanced methods, data distortion was reduced by 68.35–74.8%, and the embedding capacity of each group of watermarks increased by 6.67-43 times, respectively. It does not affect data mining classification results, and exhibits excellent robustness performance, outperforming most state-of-the-art methods.</p>