Keith T. Knox

Error-diffusion algorithm with edge enhancement

Error diffusion is a popular algorithm for the binarization of continuous-tone images. A modification of the error-diffusion algorithm is shown that introduces an input-dependent threshold into the process. This modification establishes a fast and efficient way to increase or decrease edge enhancement in the algorithm.

Automatic image annotation using adaptive color classification

We describe a system which automatically annotates images with a set of prespecified keywords, based on supervised color classification of pixels intoNprespecified classes using simple pixelwise operations. The conditional distribution of the chrominance components of pixels belonging to each class is modeled by a two-dimensional Gaussian function, where the mean vector and the covariance matrix for each class are estimated from appropriate training sets. Then, a succession of binary hypothesis tests with image-adaptive thresholds has been employed to decide whether each pixel in a given image belongs to one of the predetermined classes. To this effect, a universal decision threshold is first selected for each class based on receiver operating characteristics (ROC) curves quantifying the optimum “true positive” vs “false positive” performance on the training set. Then, a new method is introduced for adapting these thresholds to the characteristics of individual input images based on histogram cluster analysis. If a particular pixel is found to belong to more than one class, a maximuma posterioriprobability (MAP) rule is employed to resolve the ambiguity. The performance improvement obtained by the proposed adaptive hypothesis testing approach over using universal decision thresholds is demonstrated by annotating a database of 31 images.

Threshold modulation in error diffusion

A theoretical analysis of threshold modulation in error diffusion is given. It is shown that spatial modulation of the threshold is mathematically identical to processing an equivalent input image with the standard error diffusion algorithm. The equivalent input is the sum of the original image with a high-pass-filtered version of the threshold spatial modulation. The filter is a function only of the weights used to distribute the errors. This result can be used to explain the several published observed effects of threshold modulation, such as edge enhancement and the effects of adding noise to the threshold.

Image retrieval from astronomical speckle patterns

A technique has recently been proposed for processing astronomical photographs to remove the effects of turbulence. The method involves averaging over many short-exposure photographs of an object to reconstruct a diffraction-limited image of the object. A two-dimensional computer simulation of the reconstruction technique is presented. The results indicate that diffraction-limited imaging with a 100 in. telescope is possible even with 2 arc sec seeing conditions.

Multispectral imaging of the Archimedes palimpsest

Multispectral imaging techniques are being applied to improve the readability of the text in a tenth-century manuscript that includes seven treatises of Archimedes. The manuscript was erased and overwritten about 200 years later with the text of a Christian prayer book. This talk reports on the results of the multispectral imaging techniques used on the Archimedes palimpsest.

Embedding digital watermarks in halftone screens

Invisible watermark images can be incorporated in printed halftone images using specially designed halftone screens. The watermark information is embedded into the image by varying the spatial correlation of the halftone texture. The halftone screen with embedded watermarks can be used exactly as a normal halftone screen, so there is no additional image processing required for processing individual images to embed watermarks. Once the binary output image is printed on the paper, the correlation of the binary image is converted into physical spatial correlation between neighboring areas of the printed image. This correlation relation is not visible to the eye but it can be detected by scanning the printed image on a desktop scanner and processing the scanned image. Printer and scanner distortions can interfere with the self-alignment of the scanned image, so localized adjustments are made to detect the embedded spatial correlation information in the watermarked image. The retrieval of this watermark is robust to copying and distortion and it can be detected in reproductions of the halftone image.

Digital watermarks using stochastic screens

A method will be shown to incorporate digital watermarks in printed halftone images using stochastic screens. The watermark is not visible to the eye and introduces no loss in image quality. Although it cannot be seen, the watermark can be extracted at a later time with post processing. Watermarks of high contrast are incorporated in the image by varying the statistics of the stochastic screen. The watermark information can be made visible by comparing the relative changes in spatial correlation in the halftone texture of the image. Watermarking allows a printed image to be tested for the purposes of identifying the owner or the source of the image. Arbitrary customer information can be incorporated into the image, including variable information, such as the data or time of day. The technique is robust to copying of the printed image and the watermark can be detected in reproductions of the halftoned image.

Error image in error diffusion

The concept of an error image is defined for the error diffusion algorithm. Ordinarily hidden from view, the error image is a visual representation of the internally generated errors from which the algorithm derives its name. In this paper, it is shown that the error image contains a linear component of the input image, which induces edge enhancement in the output error diffusion image. Examples are shown for three different error weight distributions: a 1-D one- ahead distribution, the standard 4-element distribution, and a 12-element error distribution. The amount of edge enhancement in the corresponding algorithm is shown to vary with the amount of input image information present in the error image.

Enhancement of overwritten text in the Archimedes Palimpsest

The Archimedes Palimpsest is a thousand-year old overwritten parchment manuscript, containing several treatises by Archimedes. Eight hundred years ago, it was erased, overwritten and bound into a prayer book. In the middle of the twentieth century, a few pages were painted over with forged Byzantine icons. Today, a team of imagers, scholars and conservators is recovering and interpreting the erased Archimedes writings. Two different methods have been used to reveal the erased undertext. Spectral information is obtained by illuminating the manuscript with narrow-band light from the ultraviolet, through the visible wavebands and into the near-infrared wavelengths. Characters are extracted by combining pairs of spectral bands or by spectral unmixing techniques adapted from remote sensing. Lastly, since all of the text was written with iron gall ink, X-Ray fluorescence has been used to expose the ink underneath the painted icons. This paper describes the use of color to enhance the erased text in the processed images and to make it visible to the scholars. Special pseudocolor techniques have been developed that significantly increase the contrast of the erased text and make it possible to be read by the scholars despite the presence of the obscuring, overlaid text.

IMAGING THE DEAD SEA SCROLLS

The authors explore the current use of modern image processing techniques to decipher ancient texts, and look toward future applications, including “discovering” original texts from overwritten documents.