This paper presents a solution to recover intrinsic color features of objects for classification by using dichromatic reflection model and finite dimensional linear model for illumination and reflectance. At first the spectral power distribution of illumination is estimated. Subsequently, the surface spectral reflectance is derived with the help of the two models mentioned above. The normalized deviation spectral reflectance preserves only the contribution of body reflectance of the surface which is independent of illumination and geometric condition. Therefore, it can be used as the reliable feature for classifying color of the object. The experimental results and analysis about classification are shown in the paper.