When the surface tiny defects are illuminated by the light with different wavelengths, their optical image exist nonlinear and amplified distortion phenomenon with different degrees. It has certain effects on target extraction and signal-to-noise ratio (SNR) of the imaging information. An approach that defect image illuminated by red, green, and blue light source is used for dynamic spectral coding imaging is proposed. The image synthesis with high redundancy can be realized through pixel-level image fusion. Then it can further enhance SNR while fetching rich details of micro defects on the surface. The spectral coding theory is analyzed based on three-primary-color images, and then the dynamic weight fusion imaging method is proposed based on the image gradient. The corresponding noise analysis model of spectral nonlinear amplified signal is given. The effectiveness of dynamic weight fusion reducing standard deviation is fully verified by both theoretical analysis and numerical simulation. In the bright field surface defect imaging experiment, color complementary metal oxide silicon (CMOS) camera of three-primary-color equilibrium response is adopted to obtain three-primary-color filtering image of single micron-size defect point on the surface. By comparing with the results of traditional edge extraction, non-dynamic weight combination, and other methods, it is proved that the proposed optimization method can get a high SNR image with richer details. In the dark field surface defect imaging experiment, image gray level mean gradient and extracted defect quantity as evaluation parameters, namely two aspects of optimizing image quality and defect recognition ability, are used to futher verify the effectiveness of high fidelity and low noise. The study of this dynamic spectral coding fusion imaging method can effectively reduce the noises from spectral nonlinear amplification.