Most of the studies so far have used the green component of RGB for blood vessel recognition in eye fundus images with some using luminance. However, the suitableness of such choice and possible alternatives have not been sufficiently explored. We aimed to find the linear color combinations that would be optimal for classification of the eye fundus image pixels. It has been shown that the use of color combination optimized for "local" classification instead of green color improves results of the segmentation algorithm and retains the quality of results of tracing algorithm, while color combinations optimized for "global" and "superglobal" classification improve the sensitivity of segmentation algorithm at the price of specificity. The results indicate possibility of developing blood vessel detection algorithms that would combine advantages of different color combinations.