Robert C. Vogt

Mosaic construction, processing, and review of very large electronic micrograph composites

A system of programs is described for acquisition, mosaicking, cueing and interactive review of large-scale transmission electron micrograph composite images. This work was carried out as part of a final-phase clinical analysis study of a drug for the treatment of diabetic peripheral neuropathy. More than 500 nerve biopsy samples were prepared, digitally imaged, processed, and reviewed. For a given sample, typically 1 000 or more 1 .5 megabyte frames were acquired, for a total of between 1 and 2 gigabytes of data per sample. These frames were then automatically registered and mosaicked together into a single virtual image composite, which was subsequently used to perform automatic cueing of axons and axon clusters, as well as review and marking by qualified neuroanatomists. Statistics derived from the review process were used to evaluate the efficacy of the drug in promoting regeneration of myelinated nerve fibers. This effort demonstrates a new, entirely digital capability for doing large-scale electron micrograph studies, in which all of the relevant specimen data can be included at high magnification, as opposed to simply taking a random sample of discrete locations. It opens up the possibility of a new era in electron microscopy—one which broadens the scope of questions that this imaging modality can be used to answer.

Word-level recognition of multifont Arabic text using a feature vector matching approach

Many text recognition systems recognize text imagery at the character level and assemble words from the recognized characters. An alternative approach is to recognize text imagery at the word level, without analyzing individual characters. This approach avoids the problem of individual character segmentation, and can overcome local errors in character recognition. A word-level recognition system for machine-printed Arabic text has been implemented. Arabic is a script language, and is therefore difficult to segment at the character level. Character segmentation has been avoided by recognizing text imagery of complete words. The Arabic recognition system computes a vector of image-morphological features on a query word image. This vector is matched against a precomputed database of vectors from a lexicon of Arabic words. Vectors from the database with the highest match score are returned as hypotheses for the unknown image. Several feature vectors may be stored for each word in the database. Database feature vectors generated using multiple fonts and noise models allow the system to be tuned to its input stream. Used in conjunction with database pruning techniques, this Arabic recognition system has obtained promising word recognition rates on low-quality multifont text imagery.

A Spatially Variant, Locally Adaptive, Background Normalization Operator

This paper describes a spatially variant, locally adaptive, background normalization operator that is defined in terms of morphological openings and closings. The variable opening (and closing) residue operators address the problems that can occur when one attempts to choose a single structuring element size for background normalization purposes, in cases where the objects of interest have multiple or unknown widths. The operators attempt to assign a natural width to each pixel in an image, based on the opening or closing size that causes the largest change in its value. This size then also determines a local contrast value for that pixel. The size, local contrast and original grey values of pixels in an image can be used to cluster pixels as an aid to performing image segmentation, or foreground-background discrimination. This paper describes and illustrates the algorithm, and discusses how its outputs may be used for image analysis and parameter selection.

Morphological feature-set optimization using the genetic algorithm

This paper is an investigation into the use of genetic algorithm techniques for doing optimal feature set selection in order to discriminate large sets of characters. Human experts defined a set of over 900 features from many different classes which could be used to help discriminate different characters from a chosen character set. Each of the features was assigned a cost, based on the average amount of CPU time necessary to compute it for a typical character. The goal of the task was to find the subset of features which produced the best trade-off between recognition accuracy and computational cost. The authors were able to show that by using all of the features or even major classes of them, high rates of discrimination accuracy for a printed character set (above 98% correct, first choice) could be obtained. Application of the genetic algorithm to selected subsets of characters and features demonstrated the ability of the method to significantly reduce the computational cost of the classification system and maintain or increase accuracy from the case where a complete set of features was used.

Disambiguation and spelling correction for a neural network based character recognition system

Various approaches have been proposed over the years for using contextual and linguistic information to improve the recognition rates of existing OCR systems. However, there is an intermediate level of information that is currently underutilized for this task: confidence measures derived from the recognition system. This paper describes a high-performance recognition system that utilizes identification of field type coupled with field-level disambiguation and a spell-correction algorithm to significantly improve raw recognition outputs. This paper details the implementation of a high-accuracy machine-print character recognition system based on backpropagation neural networks. The system makes use of neural net confidences at every stage to make decisions and improve overall performance. It employs disambiguation rules and a robust spell-correction algorithm to enhance recognition. These processing techniques have led to substantial improvements of recognition rates in large scale tests on images of postal addresses.

Robust Extraction of Axon Fibers from Large-Scale Electron Micrograph Mosaics

This paper describes algorithms for extraction of axon fibers and groups of such fibers from transmission electron micrograph (TEM) image mosaics. These algorithms were developed as part of a drug evaluation study, to assess the effectiveness of a certain compound in regenerating nerve fibers of diabetic patients suffering from peripheral sensory loss due to nerve degeneration. The extraction algorithms were developed in order to pre-cue axons and potential clusters, so to reduce the workload of neuroanatomist reviewers who would otherwise be required to manually mark hundreds or thousands of such events for each biological sample. Because of the high magnification required to evaluate the regenerative clusters, typically 1000 or more electron micrographs had to be acquired, digitally registered, mosaicked, processed, and finally reviewed and marked by an anatomist, for each of more than 500 samples. The cueing algorithms described here were able to significantly reduce the workload of the reviewers, by identifying roughly 95% of the axons, with only a 1–2% false alarm rate, based on a reasonable computation time of about 1 hour per sample on a fast workstation.

Feature set optimization for the recognition of Arabic characters using genetic algorithms

This paper describes an investigation into the use of genetic algorithm techniques for selecting optimal feature sets in order to discriminate large sets of Arabic characters. Human experts defined a set of over 900 features from many different classes which could be used to help discriminate different characters from the Arabic character set. Each of the features was assigned a cost, based on the average amount of CPU time necessary to compute it for a typical character. The goal of the optimization was to find the subset of features which produced the best trade-off between recognition accuracy and computational cost. Using all of the features, or particular subsets, we obtained high recognition rates on machine-printed Arabic characters. Application of the genetic algorithm to selected subsets of characters and features demonstrates the ability of the method to significantly reduce the computational cost of the classification system and maintain or increase the recognition rate obtained with the complete set of features.

Set discrimination analysis tools for grey-level morphological operators

When considering ways to automate the generation of image processing algorithms for object recognition tasks, one critical element is the availability of measures to assess the potential and actual ability of individual operations for making a set of desired discriminations. This paper discusses the analysis and evaluation of grey-level image processing operators or algorithms from the perspective of trying to search automatically through a large space of them for one which satisfactorily performs a given recognition or discrimination task. Performance of an operator may be expressed in terms of accuracy, consistency, and cost, over an entire set of training images.The major issues of evaluating and choosing between operators in this context are discussed, and examples are given of measures which can be used to evaluate classes of operators for applicability, or to evaluate individual operators or parameter settings for actual performance. The paper first describes the form of the analysis for binary morphological operators, and then shows how it may be directly extended to grey-level morphological operators. Several examples are provided to show how grey-level pixel sets may be discriminated on the basis of various combinations of grey level and spatial criteria, as calculated by the basic morphological operators of erosion, dilation, opening, and closing.

Role of performance evaluation in automated image algorithm generation

When considering ways to automate the generation of image processing algoritms for object recognition tasks, one critical element is the availability of measures to assess the potential and actual ability of individual operations for making necessary discriminations. This paper discusses performance evaluation of image processing operators or algorithms from the perspective of trying to search automatically through a large space of them for one which satisfactorily performs a given recognition task. Performance is expressed in terms of accuracy, consistency, and cost, over a set of training images. The major issues of evaluating and choosing between operators in this context are discussed and examples are given of measures which can be used to evaluate classes of operators for applicability, as well as individual operators or parameter settings for actual performance. Examples are drawn primarily from binary morphology, with detailed extensions described for grey level morphological and linear operations.

Multiscale analysis based on mathematical morphology

In the field of computer vision, multiscale analysis has received much attention in the past decade. In particular, Gaussian scale space has been studied extensively and has proven to be effective in multiscale analysis. Recent research has shown that morphological openings or openings or closings with isotropic structuring elements such as disks define a scale space, where the radius of a disk r is the scale parameter which changes continuously from 0 to infinity. The behaviors of objects described by the morphological scale space provide strong knowledge for multiscale analysis. Based on the theory of morphological scale space, we address in this paper the two fundamental problems in multiscale analysis: (1) how to select proper scale parameters for various applications, and (2) how to integrate the information filtered at multiscales. We propose two algorithms, binary morphological multiscale analysis (BMMA) and gray-scale morphological multiscale analysis (GMMA), for extracting desired regions from binary and gray images.