Global and local document degradation models

Abstract

Two sources of document degradation are modeled -a) perspective distortion that occurs while photocopying or scanning thick, bound documents, and ii) degradation due to perturbation i n the optical scanning and digitization process: speckle, blurr, jitter, threshold. Perspective distortion is modeled by studying the underlying perspective geometry of the optical system of photocopiers and scanners. An illumination model is described to account for the non-linear intensity change occuring across a page in a perspectivedistorted document. The optical distortion process is modeled morphlogically. First a distance transform on the foreground is performed and followed by a random inversion of binary pixels where the probability of flip is a function of the distance of the pixel to the boundary of the foreground. Correlating the flipped pixels is modeled by a morphological closing operation. I IntroductionThere are many reasons for modeling document degradation. First, in order to study the performance of any OCR algorithm, it is necessary to characterize the perturbation in the output performance as a function of the perturbation in the input. This is possible oiily if we have a perturbation/degradation model for the input document. Second, a degradation model permits the evaluation of an algorithm for a continuum of degradation levels, from low degradation levels to high degradation levels. This in turn allows us to locate the 'break-down' point or the 'knee' of the algorithm which is not available in the commonly used evaluation methods, e.g., confusion matrices. Third, a knowledge of the degradation model can enable us to design algorithms for restoring degraded documents. Furthermore, OCR algorithm designers can make use of these degradation models explicitly rather than implicitly as is usually done in current literature.In this paper we model two sources of document degradation -i) perspective distortion that occurs while photocopying or scanning thick, bound documents, and ii) degradation due to perturbation in the optical process: speckle, blur, jitter, threshold, etc.. Perspective distortion is modeled by studying the underlying perspective geometry of the optical system of photocopiers and scanners. An illumination model is proposed to account for the non-linear intensity change occurring across a page in a perspectivedistorted document. The local optical distortion process is modeled morphologically. First a distance transform on the foreground is performed and followed by a random inversion of binary pixels where the probability of flip is a function of the distance of the pixel to the boundary of the foreground. Correlating the flipped pixels is modeled by a morphological closing operation.Baird [BaiSO] discusses a model for character degradation. His-model does not account for the.nonlinear distortions produced from perspective distortions. Loce [LL90] models the perturbation introduced due to mechanical disturbances in high-end Xerox photocopiers. Our paper models the distortions in geometry and illumin...

A protocol for performance evaluation of line detection algorithms

“…For the pixel-level shape-preservation evaluation the ground truth is the original raster image on which the detection algorithms operate. To evaluate the processing capability (robustness) of image analysis algorithms on real-life, noisy images, image degradation models have been proposed by Kanungo et al (1993) and used by Haralick (1992) and Hori and Doermann (1996). The ground truth image is the original clear image, and the actual input image is the degraded image.…”

A two-state Markov chain model of degraded document images

“…We use the statistical validation method proposed by Kanungo et al [9] for the validation of our noise model. The model validation procedure uses two degraded character sequences of which one is the real image sequence X = { } Step 3 Repeat Partition Repeat step 2, K number of times.…”

“…An alternative approach to modeling is the morphological document degradation model proposed by Kanungo et al [10]. Their model simulates both the statistically independent pixel inversion that occurs in images and the blurring caused by point spread function of the scanner optical system.…”

“…Their model simulates both the statistically independent pixel inversion that occurs in images and the blurring caused by point spread function of the scanner optical system. Kanungo et al [9] have proposed a statistical methodology for the validation of document image models and estimation of model parameters.…”

Functional parts detection in engineering drawings: Looking for the screws

“…This is usually done in an ad hoc manner by performing expensive trial and error runs to tune the running parameters of the algorithms. Recently, Kanungo e~ al [9] have proposed a physics-based model for the local distortions introduced by printing and scanning processes. We use this model to generate degraded synthetic data to characterize the performance of the detection algorithms and to automatically select their optimal running parameters.…”