The problem of evaluating the maximum number of information bits that can be hidden within an image is considered; usually it is addressed by looking at the watermarking process as a communication task, in which the signal, i.e. the watermark, is transmitted over the channel, the host data acts the part of. Thus the maximum number of information bits is the capacity of the watermark-channel. By relying on experimental results in which the dependence of the watermark capacity upon the watermark strength G is evidenced, the knowledge of the maximum allowed watermark level, under the constraint of watermark invisibility, is required. G is often interactively adjusted to the image at hand, because no simple algorithm exists that permits to fit the watermark level according to the characteristics of the host image. Hence, a novel algorithm to model the Human Visual System has been developed which considers frequency sensitivity, local luminance and contrast masking effects. The proposed method exploits a block based DCT decomposition of the image, that permits to trade off between spatial and frequency localisation of the image features and disturbs. Through this model, the maximum allowable watermark strength is determined in a completely automatic mode and then the value of watermark capacity is computed.