A fast algorithm for preserving noise while reducing image size

When a standard image thumbnail is generated by filtering and subsampling the original, information about the blur and noise of the original is lost. This loss of information makes it difficult to use standard thumbnails for image browsing since they do not distinguish between good quality and bad quality originals. This paper addresses image noise, developing a fast algorithm for preserving noise while reducing image size. Multirate signal transformations modify a traditional denoising and subsampling approach so that most of the processing occurs at the low spatial sampling rate of the reduced-sized output image. Experiments with noisy images and simulations show that the results of the new algorithm are very similar to the results of traditional denoising and subsampling. For the subsampling ratios typical between originals and image thumbnails, the new algorithm is an order of magnitude faster.