A proof of concept for a frost detection imager using reflected starlight is presented; the limitations of this technique are explored experimentally. An ice-covered lunar surface is simulated inside a vacuum chamber, which is then illuminated with a lamp containing UV and visible output to simulate the wavelengths of the background starfield. The simulated lunar surface is imaged with a camera utilizing a UV and visible filter pairing. At Lyman-alpha wavelengths, ice has low reflectivity, and on average appears darker than the regolith in the UV image. In visible wavelengths, this behaviour is reversed, with ice appearing brighter than regolith. UV/VIS image ratioing is subsequently performed in order to discern frost from the lunar regolith simulant in order to demonstrate the capability of this technology for locating the presence of ice on the lunar surface. When the two images are ratioed, the signal to noise ratio to distinguish ice from regolith improves by 36%. In cases where the presence of shadows and specular reflection make distinguishing ice from regolith in either a single UV or visible image difficult, ratioing the images makes the distinction clear.