By normalizing the values of its pixels with respect to the length of the used scale, a gray image can be interpreted as a fuzzy relation R which is divided in submatrices (possibly square) called blocks. Every block RB is compressed to a block GB, which in turn is decompressed to a block DB (unsigned) ⩾RB. Both GB and DB are obtained via fuzzy relation equations with continuous triangular norms in which fuzzy sets with Gaussian membership functions are used as coders. The blocks DB are recomposed in order to give a fuzzy relation D. We use the Lukasiewicz t-norm and a watermark (matrix) is embedded in every GB with the LSBM (Least Significant Bit Modification) algorithm by obtaining a block , decompressed to a block (signed). Both and are obtained by using the same fuzzy relation equations. The blocks are recomposed by obtaining the fuzzy relation (signed). By evaluating the quality of the reconstructed images via the PSNR (Peak Signal to Noise Ratio) with respect to the original image R, we show that the signed image is very similar to the unsigned image D for low values of the compression rate.