Statistical method for the prediction of matching results in spectral file searching
In file-search techniques, the distribution of mismatches quantitatively measures the fit of a particular unknown against a given library. A simple theoretical method is developed to predict, a priori, this distribution. With this procedure, for any unknown code and library, the mean number of mismatches can be calculated exactly. For a given library and coding scheme, the mismatch distribution for most unknown codes closely follows a common curve which is easily calculated from statistical properties of the library. When this common curve is normalized to a constant mean, the observed matching behavior is well predicted. The theory should permit the user to more clearly assess the effects of errors on file search performance and suggest techniques by which recognition performance may be optimized.