Generally, in watermarking techniques the size of the watermark is very small when compared to the host image. In other words, a little amount of watermark is embedded in the huge quantity of image pixels as the notice of legitimate ownership. Contrary to that idea, this is an attempt in which the capacity of watermarking is improved by embedding huge amount of watermark efficiently in the less quantity of image pixels. The core idea behind the proposed approach is to select watermarkable pixels from the host image based on the census transform and hamming distance followed by the embedding which is done by proposed spectral decompositions, i.e., Hankel, Circulant and Topelitz spectral decomposition. Finally, a reliable watermark extraction scheme is developed which is free from the false-positive detection problem of singular values. The experimental evaluation demonstrates that the proposed scheme is expeditiously able to withstand a variety of extreme attacks and highly suitable for covert communications. © 2016, Springer Science+Business Media New York.