An improved data hiding in halftone images with cooperating pair toggling human visual system (PTHVS) is presented in this paper. An objective halftone image quality evaluation method based on the human visual system obtained by least-mean-square (LMS) is also introduced. By rigorously searching the optimum toggled pixels with the proposed human visual LMS-trained filter, the proposed method is proven to be superior to the data hiding smart pair toggling (DHSPT), proposed by Fu and Au, in image quality under a number of tested halftone images. The tested halftone images include ordered dithering, Floyd error diffusion, Jarvis error diffusion, and Stucki error diffusion images. Moreover, the proposed method offers high embedded capacity, and it is flexible to deal with different capacity applications