A novel radiometric compensation algorithm is proposed to correct the color distortions of the projection display on patterned surface. Two choices of the objective image are investigated in this algorithm. A linear compression followed by a sigmoid-function transformation based on the statistical parameters of the compensating radiance map is performed to reduce the clippings of the compensating color and expected to properly reproduce the dynamic range and the visibility of the image content. Three types of projection displays are produced for each test image. The parameter analysis indicates that the compensated displays could well conceal the surface pattern and tend to exhibit higher lightness or lightness contrast in the camera. The comparison results in a psychophysical experiment demonstrate that all the compensated displays outperform the uncompensated ones. The original input images seem more appropriate to be selected as the objective images, and the predicted objectives on the white screen are just proper for the bright and low-contrast input images displayed on the faint surface.