The general problem that we consider is the """"""""real-time"""""""" interpretation of a noisy image as it arrives column by column. The interpretation is modeled by a binary image of the same size. We assume that there is a statistical distribution (that depends on the application area), which assigns to each such binary image a probability of occurrence. The usefulness of this model depends on being able to extract with the help of the binary image the essence of the information that the sender of the original image intended to convey. Assuming such a model, our task is to design an algorithm that will produce the columns of the sought-after binary image as the noisy image arrives column by column. The algorithm should make use of the statistics associated with the binary images and with the noise in the received images. We plan to test the applicability of such algorithms for noise reduction in hearing aids. Signals are turned into images by taking a transform such as the short-time Fourier transform (STFT) of the signal. These images are extremely redundant. It is our hypothesis that, for every such image, there exist a binary image that contains all the essential speech information. This will be tested using normal-hearing volunteers.
