Methods of the steganography are characterized by such efficiency rates as invisibility, robustness and capacity. There is considered the maximum capacity support of the information embedding into the DCT-domain. It is investigated the known algorithm that realizes the adaptive information embedding into the digital images frequency domain. The adaptivity is reached due to the image partition into the unequal blocks using a quad-tree. There is received the improved modification of the algorithm based on the reference point variation in case of the image partition into the blocks. The received modification allows to provide the better invisibility at the same capacity.