A variety of color calibration technologies have been developed for input and output devices. Linear or nonlinear matrices have been conveniently applied to correct the color filter's mismatch with color matching function in scanners or suppress the crosstalk by unwanted absorption of colorants in printers. The color matching accuracy is expected to be further improved when the nonlinear matrices are optimized into subdivided smaller color spaces than in single matrix of the entire color space. This article proposes a new method for partitioning the color space into sub-spaces divided by the combinations of luminance, chrominance, radius, or hue angle in CIELAB space so that each sub space includes the constant number of color samples. Linear or nonlinear color correction functions are applied to each subdivided space and the coefficient matrices are optimized individually by the method of least squares. The new method resulted in the high precision color matching with rms color differences ΔE_ab^* (rms) <0.5 for flat bed scanner and ΔE_ab^*(rms) ≈ 2.0 for inkjet printer. The color matching accuracies could approach the colorimetric measurement errors in scanners and mechanical stabilities in printers by the proposed subspace division methods.