The very-large-scale integration (VLSI) implementation of a probabilistic neural network image interpolator based on a neural network model is provided in this paper. The interpolator takes into consideration both smoothness (flat region) and sharpness (edge region) characteristics at the same model. A single neuron, combined with particle swarm optimization training, is used for sharpness/smoothness adaptation. A highly efficient VLSI architecture of a probabilistic neural network image interpolator is designed in field-programmable gate array for supporting real-time digital zooming/scaling applications. The functional architecture of the proposed image interpolator is decomposed in three main functional modules: edge adaptation module, Gaussian module and interpolator module, as well as a top-level pipelining controller. The data throughputs achieved are 20 fps in XGA format at 20 Mhz system clock.