Abstract: Color-to-gray conversion for digital color images is widely used in many applications. In this paper we propose an efficient gradient domain color-to-gray conversion algorithm depending on automatic optimization of parameters. A gradient field, defined with the luminance gradient and a modulated chromatic difference enhancement in CIELAB space, is created to construct the grayscale image using a Poisson Equation Solver (PES). In order to distinguish isoluminant colors, we define a sign function for the gradient field to keep correct color ordering. In the inefficient preprocess step, the four parameters of this method are automatically optimized in the sense of human vision with a structural similarity index measurement (SSIM). Since the optimal values of parameters β, γ and α are similar for different images, we set them as empirical optimal values, and the remaining parameter θ is automatically optimized following another efficient heuristic linear separation rule. Experimental results show that our algorithm is efficient to produce perfect grayscale images which have properties of salience preserving, color discrimination and coinciding with human perception to color difference.