A comprehensive metal surface reflection model is introduced to demonstrate the feasibility of the differential visual measurement theory. In order to seek the factual basis, the reflection brightness of some roughness comparison specimens is measured in the plane of incidence, and then the distribution law is summarized based on a large amount of data. A vision system is built, and the images collected from the bright and dark fields are fused by differential algorithm. Experimental results show that the target surface defects are greatly enhanced and the clutter background information is effectively suppressed. The differential visual measurement system based on the reflective properties of the metal surface shows its superiority in improving contrast and enhancing edge of images.