Eugene H. Ratzlaff

A tandem-lens epifluorescence macroscope: Hundred-fold brightness advantage for wide-field imaging

The design of a macroscope constructed with photography lenses is described and several applications are demonstrated. The macroscope incorporates epi-illumination, a 0.4 numerical aperture, and a 40 mm working distance for imaging wide fields in the range of 1.5-20 mm in diameter. At magnifications of 1X to 2.5X, fluorescence images acquired with the macroscope were 100-700 times brighter than those obtained with commercial microscope objectives at similar magnifications. In several biological applications, the improved light collection efficiency (20-fold, typical) not only minimized bleaching effects, but, in concert with improved illumination throughput (15-fold, typical), significantly enhanced object visibility as well. Reduced phototoxicity and increased signal-to-noise ratios were observed in the in vivo real-time optical imaging of cortical activity using voltage-sensitive dyes. Furthermore, the macroscope has a depth of field which is 5-10 times thinner than that of a conventional low-power microscope. This shallow depth of field has facilitated the imaging of cortical architecture based on activity-dependent intrinsic cortical signals in the living primate brain. In these reflection measurements large artifacts from the surface blood vessels, which were observed with conventional lenses, were eliminated with the macroscope.

Methods, reports and survey for the comparison of diverse isolated character recognition results on the UNIPEN database

A framework of data organization methods and corresponding recognition results for UNIPEN databases is presented to enable the comparison of recognition results from different isolated character recognizers. A reproducible method for splitting the Train-R01/V07 data into an array of multi-writer and omni-writer training and testing pairs is proposed. Recognition results and uncertainties are provided for each pair, as well as results for the DevTest-R01/V02 character subsets, using an online scanning n-tuple recognizer. Several other published results are surveyed within this context. In sum, this report provides the reader multiple points of reference useful for comparing a number of published recognition results and a proposed framework that similarly allows private evaluation of unpublished recognition results.

A scanning n-tuple classifier for online recognition of handwritten digits

A scanning n-tuple classifier is applied to the task of recognizing online handwritten isolated digits. Various aspects of preprocessing, feature extraction, training and application of the scanning n-tuple method are examined. These include: distortion transformations of training data, test data perturbations, variations in bitmap generation and scaling, chain code extraction and concatenation, various static and dynamic features, and scanning n-tuple combinations. Results are reported for both the UNIPEN Train-R01/V07 and DevTest-R01/V02 subset la isolated digits databases.

A workstation interface for measuring interspike intervals

A method is presented for collecting and storing, singly or in parallel, interspike intervals to a DOS-based workstation as a background process. The method comprises digital electronics and low-level software effecting an interface mechanism from the receipt of digital spike pulses to the recording of user-accessible data arrays readily manipulable by high-level software. Functional details, requirements, and application to parital cortex single unit physiology are discussed.

Probability table compression for handwritten character recognition

This paper presents a new probability table memory compression method based on mixture models and its application to N-tuple recognizers and N-gram character language models. Joint probability tables are decomposed into lower dimensional probability components and their mixtures. The maximum likelihood parameters of the mixture models are trained by the expectation maximization (EM) algorithm and quantized to one byte integers. Probability elements that mixture models do not estimate reliably are kept separately. Experimental results with on-line handwritten UNIPEN uppercase and lowercase characters show that the total memory size of an on-line scanning N-tuple recognizer is reduced from 12.3 MB to 0.66 MB bytes, while the recognition rate drops from 91.64% to 91.13% for uppercase characters and from 88.44% to 87.31% for lowercase characters. The N-gram character language model was compressed from 73.6 MB to 0.58 MB with minimal reduction in performance.

EM mixture model probability table compression

This paper presents a new probability table compression method based on mixture models applied to N-tuple recognizers. Joint probability tables are modeled by lower dimensional probability mixtures and their mixture coefficients. The maximum likelihood parameters of the mixture models are trained by the expectation-maximization (EM) algorithm and quantized to one byte integers. The probability elements which mixture models do not estimate reliably are kept separately. Experimental results with on-line handwritten UNIPEN digits show that the total memory size of an N-tuple recognizer is reduced from 11.8 Mbytes to 0.55 Mbytes, while the recognition rate drops from 97.7% to 97.5%.

Recurrent genetic programming

A typical pattern recognition system consists of two stages: the pre-processing stage to extract features from the data, and the classification stage to assign the feature vector a class label. There are two kinds of feature extraction techniques with respect to the kind of data: the fixed number of features per sample generating a fixed length feature vector, and the fixed number of features per subsample generating a variable length feature vector due to variable number of sub-samples (frames) for each input pattern. The first kind is the most commonly used feature vector for classification methods. The second kind is usually extracted in domains where the input sample is time-variant. Traditionally a separate class of machine learning algorithms consisting of hidden Markov models, recurrent neural networks, etc. have been employed for classification of time variant signals. Evolutionary computation techniques like genetic algorithms and genetic programming have also been used previously to optimize the architecture for HMMs or learning the weights for recurrent-neural networks. We describe a recurrent framework for genetic programming (GP). This framework helps place GP in the class of machine learning algorithms alongside recurrent neural networks and hidden Markov models. We describe the application of recurrent genetic programming for the classification of on-line handwritten numerals obtained from tablet-based input.