Robert M. Haralick

Document zone content classification and its performance evaluation

This paper describes an algorithm for the determination of zone content type of a given zone within a document image. We take a statistical based approach and represent each zone with 25 dimensional feature vectors. An optimized decision tree classifier is used to classify each zone into one of nine zone content classes. A performance evaluation protocol is proposed. The training and testing data sets include a total of 24,177 zones from the University of Washington English Document Image database III. The algorithm accuracy is 98.45% with a mean false alarm rate of 0.50%.

Estimating piecewise-smooth optical flow with global matching and graduated optimization

This paper presents a new method for estimating piecewise-smooth optical flow. We propose a global optimization formulation with three-frame matching and local variation and develop an efficient technique to minimize the resultant global energy. This technique takes advantage of local gradient, global gradient, and global matching methods and alleviates their limitations. Experiments on various synthetic and real data show that this method achieves highly competitive accuracy.

Optimal matching problem in detection and recognition performance evaluation

Abstract This paper proposes a principle of one-to-one correspondence in performance evaluation of a general class of detection and recognition algorithms. Such a correspondence between ground-truth entities and algorithm declared entities is essential in accurately computing objective performance measures such as the detection, recognition, and false alarm rates. We mathematically define the correspondence by formulating a combinatorial optimal matching problem. In addition to evaluating detection performance, this methodology is also capable of evaluating recognition performance. Our study shows that the proposed principle for detection performance evaluation is simple, general and mathematically sound. The derived performance evaluation technique is widely applicable, precise, consistent and efficient.

An optimization methodology for document structure extraction on Latin character documents

In this paper, we give a formal definition of a document image structure representation, and formulate document image structure extraction as a partitioning problem: finding an optimal solution partitioning the set of glyphs of an input document image into a hierarchical tree structure where entities within the hierarchy at each level have similar physical properties and compatible semantic labels. We present a unified methodology that is applicable to construction of document structures at different hierarchical levels. An iterative, relaxation-like method is used to find a partitioning solution that maximizes the probability of the extracted structure. All the probabilities used in the partitioning process are estimated from an extensive training set of various kinds of measurements among the entities within the hierarchy. The offline probabilities estimated in the training then drive all decisions in the online document structure extraction. We have implemented a text line extraction algorithm using this framework.

Table structure understanding and its performance evaluation

This paper presents a table structure understanding algorithm designed using optimization methods. The algorithm is probability based, where the probabilities are estimated from geometric measurements made on the various entities in a large training set. The methodology includes a global parameter optimization scheme, a novel automatic table ground truth generation system and a table structure understanding performance evaluation protocol. With a document data set having 518 table and 10,934 cell entities, it performed at the 96.76% accuracy rate on the cell level and 98.32% accuracy rate on the table level.

Towards a formal concept analysis approach to exploring communities on the world wide web

An interesting problem associated with the World Wide Web (Web) is the definition and delineation of so called Web communities. The Web can be characterized as a directed graph whose nodes represent Web pages and whose edges represent hyperlinks. An authority is a page that is linked to by high quality hubs, while a hub is a page that links to high quality authorities. A Web community is a highly interconnected aggregate of hubs and authorities. We define a community core to be a maximally connected bipartite subgraph of the Web graph. We observe that the web subgraph can be viewed as a formal context and that web communities can be modeled by formal concepts. Additionally, the notions of hub and authority are captured by the extent and intent, respectively, of a concept. Though Formal Concept Analysis (FCA) has previously been applied to the Web, none of the FCA based approaches that we are aware of consider the link structure of the Web pages. We utilize notions from FCA to explore the community structure of the Web graph. We discuss the problem of utilizing this structure to locate and organize communities in the form of a knowledge base built from the resulting concept lattice and discuss methods to reduce the complexity of the knowledge base by coalescing similar Web communities. We present preliminary experimental results obtained from real Web data that demonstrate the usefulness of FCA for improving Web search.

Algorithm performance contest

This contest involved the running and evaluation of computer vision and pattern recognition techniques on different data sets with known groundwidth. The contest included three areas; binary shape recognition, symbol recognition and image flow estimation. A package was made available for each area. Each package contained either real images with manual groundtruth or programs to generate data sets of ideal as well as noisy images with known groundtruth. They also contained programs to evaluate the results of an algorithm according to the given groundtruth. These evaluation criteria included the generation of confusion matrices, computation of the misdetection and false alarm rates and other performance measures suitable for the problems. The paper summarizes the data generation for each area and experimental results for a total of six participating algorithms.

Integrated surface model optimization for freehand three-dimensional echocardiography

The major obstacle of three-dimensional (3-D) echocardiography is that the ultrasound image quality is too low to reliably detect features locally. Almost all available surface-finding algorithms depend on decent quality boundaries to get satisfactory surface models. We formulate the surface model optimization problem in a Bayesian framework, such that the inference made about a surface model is based on the integration of both the low-level image evidence and the high-level prior shape knowledge through a pixel class prediction mechanism. We model the probability of pixel classes instead of making explicit decisions about them. Therefore, we avoid the unreliable edge detection or image segmentation problem and the pixel correspondence problem. An optimal surface model best explains the observed images such that the posterior probability of the surface model for the observed images is maximized. The pixel feature vector as the image evidence includes several parameters such as the smoothed grayscale value and the minimal second directional derivative. Statistically, we describe the feature vector by the pixel appearance probability model obtained by a nonparametric optimal quantization technique. Qualitatively, we display the imaging plane intersections of the optimized surface models together with those of the ground-truth surfaces reconstructed from manual delineations. Quantitatively, we measure the projection distance error between the optimized and the ground-truth surfaces. In our experiment, we use 20 studies to obtain the probability models offline. The prior shape knowledge is represented by a catalog of 86 left ventricle surface models. In another set of 25 test studies, the average epicardial and endocardial surface projection distance errors are 3.2 /spl plusmn/ 0.85 mm and 2.6 /spl plusmn/ 0.78 mm, respectively.

Performance Evaluation of Document Structure Extraction Algorithms

This paper presents a performance metric for the document structure extraction algorithms by finding the correspondences between detected entities and ground truth. We describe a method for determining an algorithms optimal tuning parameters. We evaluate a group of document layout analysis algorithms on 1600 images from the UW-III Document Image Database, and the quantitative performance measures in terms of the rates of correct, miss, false, merging, splitting, and spurious detections are reported.

Nonlinear Manifold Clustering By Dimensionality

Because of variable dependence, high dimensional data typically have much lower intrinsic dimensionality than the number of its variables. Hence high dimensional data can be expected to lie in (nonlinear) lower dimensional manifold. In this paper, we describe a nonlinear manifold clustering algorithm. By connecting data vectors with their neighbors in feature space, we construct a neighborhood graph from given set data vectors. Furthermore, geometrical invariance, namely dimensionality, are extracted from the neighborhood of vectors, and used to facilitate the clustering procedure. In addition, we discuss a latent model for data cluster descriptions and an EM algorithm to find such descriptions. Preliminary experiments illustrate that this new algorithm can be used to explore the nonlinear structure of data