To improve the quality of the reconstructed image in any tamper condition, this work proposes a self-embedding watermarking scheme with robustness against watermark alterations, and discusses the reasonability of the predefined threshold and the reliability of tamper detection. The proposed scheme firstly sets the least significant bit (LSB) of three-quarter pixels and two LSBs of the residual pixels in the original image to zero. And then the low-frequency feature image is obtained by quantizing the low-frequency wavelet coefficients of the original image content. The improved security watermark, which is the binary code of the scramble version of the low-frequency feature image, is embedded into the LSBs which were set to zero. While the image authentication, the proposed method is able to discriminate the malicious modifications from the mild distortions according to the predefined threshold to enhance the robustness against innocuous alterations such as watermark changes and channel noise. Theoretical analysis and simulation results show that the proposed scheme can discriminate the different modifications according to the predefined threshold no matter the embedded watermark in the host image is randomly or regionally tampered. The quality of the recovery image can be effectively improved due to the fact that the different reconstructed methods are adopted for the different tamper blocks.