Normalization (division) versus subtraction of digitized images

The propagation of noise in digitized images arrived at by algebraic manipulation of two images has been investigated. The algebraic manipulations investigated were logarithmic subtraction, linear subtraction, and division of the two images. The two images were identical except for the presence of a contrast-adding component in one of them. The two extreme cases of video-fluoroscope system behavior, gray level proportional to incident photon flux and gray level proportional to the product of thickness and linear attenuation coefficient, were investigated. The ratio of signal to noise was calculated for each of these cases and a figure of merit established for the ratio of signal to noise of the division process versus each of the subtraction processes for each of the two fluoroscopic system behaviors. In all cases, this figure of merit is greater than unity. Values of these figures of merit for iodine concentrations of 5, 20, and 300 mg/cm3 are given in the paper. The process of division appears to be advantageous as a signal-to-noise enhancer for all applications not requiring a quantitative relationship between image contrast and photon flux.