Reflected light can be decomposed into specular and diffuse components according to the light reflectance theory and experiments. The specular component appears in smooth surfaces mainly, while the diffuse one is visible in rough surfaces mostly. Therefore, each component can be used in forming their correlations to a surface roughness. However, they cannot represent the whole surface roughness seamlessly, because each formulation is merely validated in their available surface roughness regions. To solve this problem, new approaches to properly blend two light components in all regions are proposed in this paper. First is the weighting function method that a blending zone and rate can be flexibly adjusted, and second is the neural network method based on the learning from the measurement data. Simulations based on the light reflectance theory were conducted to examine its performance, and then experiments conducted to prove the enhancement of the measurement accuracy and reliability through the whole surface roughness regions.