Now let us discuss the problem of the algorithm in Ref. ^{1}. It is clear to see that the most significant step in this algorithm of watermark detection is 7. Moreover, please note that the matrices $Uw$, $SWk$, $Vw$ are required in watermark extraction, and $Uw$, $Vw$ are the side information in Ganic and Eskicioglu^{1} algorithm. Hence, no matter what kind of the attacked cover image $A*$, even one whose content is not the same as that of the original $A$, is selected, but the same private matrices $Uw$, $Vw$ that are tightly connected with the original watermark $W$ are used to recover the watermark image. If we use four random matrices $S**k$, $k=1,2,3,4$, with $\lambda **>0,i=1,\u2026,n$ conducted as Eq. 7, then four matrices $W**k=UWS**kVWT$, $k=1,2,3,4$ will be generated. But $W**k$ are visual and geometrically similar to the original $W$ with overwhelming probability, because all the geometrical features of the watermark image $W$ are contained in $Uw$, $Vw$. Moreover, according to the watermark embedding procedure (4), only the singular value vectors $SW$ of watermark $W$ are inserted, while the geometrical features of the watermark image $W$ are not inserted. Hence, it is worthwhile to point out that not only is the watermark extraction algorithm problematic, but so is the watermark embedding algorithm. The intrinsic reasons for the serious problem with high probability of false positive answers are (i) the basis space of SVD is image content dependent, and (ii) there is no one-to-one correspondence between the singular value vector and watermark image.