Various embodiments of the present disclosure provide techniques for performing video denoising (VDN). An adaptive noise threshold is dynamically determined and used to distinguish between frame to frame differences in pixel values that relate to image motion from those differences that relate to noise. The disclosed techniques enable the noise threshold to be continuously updated, for example as frequently as once per frame, so that the noise threshold may closely track to varying levels of noise in the input video data. The techniques may be implemented in, for example, a video format conversion apparatus. Advantageously, the techniques may be incorporated in programmable logic devices (PLD's) or Field Programmable Gate Arrays (FPGA's) configurable to perform video format conversion, while adding only modest additional computational demands on the apparatus.