We implement a binary differential joint-transform correlator for real-time single- and multiple-target recognition applications. In a real-time situation the input scene is captured using a CCD or thermal camera. The obtained joint power spectrum is first differentiated and then binarized. The subset median threshold method was used to get the threshold value for binarization. The binarized joint power spectrum is displayed over a ferroelectric-liquid-crystal spatial light modulator and is Fourier-transformed optically to obtain the correlation peaks. Differential processing of the joint power spectrum removes the zero-order spectra (dc) and hence improves the detection efficiency. Experiments taking single as well as multiple input images have been performed. The parameters for a performance measure have also been calculated. Single and multiple targets with added Gaussian noise have also been used to check the correlation outputs. Computer simulation and experimental results are presented.