2.5D human face is illumination invariant which has a great advantage in face recognition. However, the existing method are linear based and capturing a 2.5D human face involves multi images from the same view point which is impractical in reality. This paper introduces a new nonlinear method for normal Surveillance camera to capture a 2.5D human face data. Only a single image is needed during capturing process by using RGB light sources. The illumination is separated from 2D images by applying ICA (Independent component analysis) method. A nonlinear statistical reflection model is developed through the nonlinear ICA algorithm to compensate nonlinear distortions during image capturing process. The proposed algorithm has achieved excellent features in separating the illumination which yielded very high accuracy of 2.5D human face data recovery.