This work presents a semi-blind watermark scheme exploiting a self-reference image for digital image protection and authenticity. The proposed scheme is feasible for practical applications since extraction and detection procedures do not require the original image. To improve robustness and reduce the computational cost of embedding or extracting the watermark, in the proposed scheme, the original image is initially transfromed into wavelet coefficients using one-level discrete wavelet transform (DWT) and then three high-frequency sub-bands, i.e., LH1, HL1 and HH1, are set as zero. A corresponding self-reference image can thus be obtained after performing the inversed DWT. The watermark is embedded into a specific difference value between the original image and the self-reference image by using just noticeable distortion (JND) approach. Experimental results indicate that the proposed scheme provides the high visual quality of the watermarked image and leads to satisfactory robustness to various attacks, including cropping, blurriness, brightness, noise interference, sharpening, and JEPG lossy compression. Moreover, the proposed scheme enhances the security capability of Liu et al.'s scheme to resist the attack with Ting et al's scheme.