This article proposes a new method of natural color reproduction by adaptive color conversion of an image with highlights under arbitrary illuminants. This method uses a digitized RGB value of a color image under various light sources. Each pixel of the color image is represented as a 3-dimensional vector with red, green, and blue light-value components. The unknown light source used in the capturing process of an image is then estimated as the mean vector of the spatial means of the red, green, and blue image components. In contrast, standard light sources, such as D65 or D50, that appear white to the human eye, are described as a unit vector in RGB color space if color constancy is assumed. The proposed vector transformation can thus be defined as a vector rotation, through which the white point of an estimated light source is transformed to one of standard light. Next, a 3 × 3 homogeneous transformation matrix relative to the equivalent axis is produced and the input color of each pixel is transformed into a new RGB color space by using this matrix as a color filter. However, if there are highlights in the input image, the initial method proposed can cause some distortion due to the over-transformation of highlights. Accordingly, to solve this problem this article proposes a modified transformation function where the initial transformation matrix is corrected by regression based on identifying a highlight mapping without distortion. Thereafter, the correctly transformed red, green, and blue color values of each pixel in the new space are non-linearly mapped into integer values between 0 and 255 for a 24-bit natural color display and stored in digital media. The proposed method can be used in the color conversion of an image under arbitrary illuminants to one under daylight through improving the effect of the incident illumination and thereby enhancing the image quality.