Dual energy x-ray CT images are computed using either image or projection data. The latter is thought to be preferable for two-material decomposition. Nonetheless, using effective energies of polychromatic x-ray beams at separated kVp values, material decomposition and pseudo-monochromatic reconstruction can be performed from reconstructed images. This image-based approach generates added noise which should benefit from applying processing for noise reduction. A set of material attenuation information so produced defines a vector space, which represents the true material property but is predefined from mass attenuation coefficients of major body-composite materials. We assumed 53keV and 72keV x-ray effective energies for 80kVp and 140kVp dual energy CT. The Gram-Schmidt process was applied to remove noise orthogonal to the vector spaces of the body-composite materials. Two-material decomposition was performed, and monochromatic and density images were reconstructed. Evaluations of image noise, Hounsfield unit accuracy, and resolution with a phantom, as well as with abdominal images, demonstrates improved CNR and SNR without loss of detail. This method of noise suppression also produced high quality density maps of two basis materials. Since dual energy CT currently uses slightly above average radiation dose, this method has the potential for lowering dose in addition to improving image quality.