Binarizing Low Resolution Images.

In optical character recognition (OCR), it is generally necessary to first binarize the document image. In this binarization process, the image is reduced to a one bit quantization in which each pixel is either black or white. In processing a color image, binarization is a non trivial operation since each pixel may be encoded by 24 or more bits (8 million or more colors / intensity levels). If the document image is low resolution and is encoded with a lossy compression method, such as JPEG, the problem is further compounded since lossy compression results in quantization noise or error in the decoded image. We address the problem of recovery of a high resolution binary image of low resolution color images containing text and encoded by lossy compression methods, such as JPEG. © 2006 American Institute of Physics [ABSTRACT FROM AUTHOR]