Subband coding is a powerful means for highly efficient image compression. In order to improve the coding performance of subband image coding, we recently have proposed the optimum space-frequency partition coder (OSFP) that optimizes the following three factors in the rate-distortion sense: the frequency band partition with a small number of subbands, quantization and the spatial segmentation to exclude redundant pixels. However, an encoded image obtained by OSFP is not necessarily optimal in subjective image quality because the three factors are optimized to minimize the mean square error (MSE). In this paper, we present a new OSFP that obtains a high quality coded image subjectively by optimizing the three factors so that MSE weighted by considering both the human visual sensitivity and a region-of-interest of human is minimized. Experimental results show that the quality of encoded images obtained by the proposed method has higher subjectively than them of both the conventional OSFP and JPEG2000 by the mean opinion score (MOS).