Optical pattern recognition techniques based on the optical joint transform correlator (JTC) scheme are attractive due to their simplicity. Recent improvements in spatial light modulators (SLM) increased the popularity of the JTC, providing means for real time operation. Using a binary SLM for the display of the Fourier spectrum, first requires binarization of the joint power spectrum distribution. Although hard-clipping is the simplest and most common binarization method used, we suggest to apply error-diffusion as an improved binarization technique. The performance of a binary JTC, whose input image is considered to contain additive zero-mean white Gaussian noise, is investigated. Various ways for nonlinearly modifying the joint power spectrum prior to the binarization step, which is based on either error-diffusion or hard-clipping techniques, are discussed. These nonlinear modifications aim at increasing the contrast of the interference fringes at the joint power spectrum plane, leading to better definition of the correlation signal. Mathematical analysis, computer simulations and experimental results are presented.