Ramkumar Narayanswamy

Extending the imaging volume for biometric iris recognition

The use of the human iris as a biometric has recently attracted significant interest in the area of security applications. The need to capture an iris without active user cooperation places demands on the optical system. Unlike a traditional optical design, in which a large imaging volume is traded off for diminished imaging resolution and capacity for collecting light, Wavefront Coded imaging is a computational imaging technology capable of expanding the imaging volume while maintaining an accurate and robust iris identification capability. We apply Wavefront Coded imaging to extend the imaging volume of the iris recognition application.

Signal-to-noise analysis of task-based imaging systems with defocus

We analyze the signal-to-noise ratio (SNR) of arbitrary imaging systems in the presence of defocus. The modulation transfer function (MTF) and the mean SNR are combined to calculate the spatial-frequency spectrum of the SNR (the spectral SNR). Computational imaging methods are used for extending the depth of field (DOF) of the system. The DOF of a task-specific imaging system is defined as the range of defocus that causes the spectral SNR to drop below a minimum value within a band of spatial frequencies of interest. We introduce the polar-SNR plot as a tool for visualizing the spectral SNR of defocused imaging systems with asymmetric pupil functions. As an example, we perform the analysis of an imaging system used for biometric iris recognition.

Intelligent data elimination for a rare event application

Rare event applications are characterized by the event-of- interest being hidden in a large volume of routine data. The key to success in such situations is the development of a cascade of data elimination strategies, such that each stage enriches the probability of finding the event amidst the data retained for further processing. Automated detection of aberrant cells in cervical smear slides is an example of a rare event problem. Each slide can amount to 2.5 gigabytes of raw data and only 1 in 20 slides are abnormal. In this paper we examine the use of template matching, artificial neural networks, integrated optical density and morphological processing as algorithms for the first data elimination stage. Based on the experience gained, we develop a successful strategy with improves the overall event probability in the retained data from 0.01 initially to 0.87 after the second stage of processing.

Design and optimization of the cubic-phase pupil for the extension of the depth of field of task-based imaging systems

In this article we study the use of the cubic-phase pupil function for extension of the depth of field of task-based imaging systems. In task-based design problems the resolution of interest varies as a function of the object distance due to change in magnification. This introduces a new challenge in the design process. We discuss how the optimal design criterion of task-based imaging systems is fundamentally different from that of visual imaging systems and formulate the optimization problem. We discuss how the use of the cubic-phase pupil function changes the spectral signal-to-noise-ratio (SNR) and modulation transfer function (MTF) in the range of the depth of field in order to fulfill our design requirements. We introduce an approximation to the problem of maximizing SNR and show that it is amenable to analytic treatment. We derive an explicit expression for the optimized cubic-phase pupil function parameters for a general problem of this class, thus establishing an upper bound for the extension of the depth of field using cubic-phase Wavefront Coding.

Optoelectronic hit/miss transform for screening cervical smear slides.

An optoelectronic morphological processor for detecting regions of interest (abnormal cells) on a cervical smear slide using the hit/miss transform is presented. Computer simulation of the algorithm tested on 184 Papsmear images provided 95% detection and 5% false alarm. An optoelectronic implementation of the hit /miss transform is presented, along with preliminary experimental results.

Optoelectronic region of interest detection: an application in automated cytology.

Diagnostic cytology, which is used to screen for cervical cancer, involves characterizing cellular features such as shape, size, and texture. Automated screening of cervical smear slides is desirable but computationally challenging since each slide requires processing 2 x 10(9) pixels at a resolution of 0.8 mum per pixel. We demonstrate that the throughput of optical processors can be exploited in automated cervical smear-screening systems. In particular, we identify a morphological shape detector to perform the initial region of interest (ROI) detection and to demonstrate experimentally its optoelectronic implementation. The ROI detector is tested on 200 images, and its performance is characterized as a receiver operating characteristic (ROC). The area under the ROC curve is as high as 96.4% of the total area. The simulation and the experimental results are found comparable, and the discrepancy between the two results is determined to be a function of the number of bits represented in the filter plane device.

Morphological feature detection for cervical cancer screening

An optoelectronic system has been designed to pre-screen pap-smear slides and detect the suspicious cells using the hit/miss transform. Computer simulation of the algorithm tested on 184 pap-smear images detected 95% of the suspicious region as suspect while tagging just 5% of the normal regions as suspect. An optoelectronic implementation of the hit/miss transform using a 4f Vander-Lugt correlator architecture is proposed and demonstrated with experimental results.

Optoelectronic region of interest detection in monolayer cervical smear slides

An optoelectronic detection system using two electrically addressed spatial light modulators in an optical correlator has been constructed to find regions of interest in cervical smear slides using the hit/miss transform algorithm. The purpose of the detector is to locate abnormal cells in the cervical smear and mark the region of interest for further classification by a second stage to the overall system. In addition, an image database of characteristic monolayer cervical smear images has been constructed for testing the system. The optoelectronic processing of cytological specimens can in theory provide both an improvement in the speed of scanning a slide for a region of interest and also a decrease in current manual screening errors. Results of the optoelectronic correlator and corresponding computer simulations will be discussed as well as further means of improving the system. Conclusions about further steps in the implementation of a complete medical diagnostic system including classification of regions of interest and improvements for automation will also be addressed.

Optoelectronic image processing for cervical cancer screening

Automation of the Pap-smear cervical screening method is highly desirable as it relieves tedium for the human operators, reduces cost and should increase accuracy and provide repeatability. We present here the design for a high-throughput optoelectronic system which forms the first stage of a two stage system to automate pap-smear screening. We use a mathematical morphological technique called the hit-or-miss transform to identify the suspicious areas on a pap-smear slide. This algorithm is implemented using a VanderLugt architecture and a time-sequential ANDing smart pixel array.