Robert T. Gray

A fully automatic redeye detection and correction algorithm

A fully automatic redeye detection and correction algorithm was developed at Eastman Kodak Company Research Laboratories. The algorithm is highly sophisticated so that it is able to distinguish most redeye pairs from scene content. It is also highly optimized for execution speed and memory usage enabling it to be included in a variety of products. Detected redeyes are corrected so that the red color is removed, but the eye maintains a natural look.

A computational approach to determination of main subject regions in photographic images

Abstract We present a computational approach to main subject detection, which provides a measure of saliency or importance for different regions that are associated with different subjects in an image with unconstrained scene content. It is built primarily upon selected image semantics, with low-level vision features also contributing to the decision. The algorithm consists of region segmentation, perceptual grouping, feature extraction, and probabilistic reasoning. To accommodate the inherent ambiguity in the problem as reflected by the ground truth (probabilistic in nature), we have developed a novel training mechanism for Bayes nets based on fractional frequency counting. Using a set of images spanning the ‘photo space,’ experimental results have shown the promise of our approach in that most of the regions that independent observers ranked as the main subject are also labeled as such by our system. In addition, without reorganization and retraining, the Bayes net-based framework lends itself to performance scalable configurations to suit different applications that have different requirements of accuracy and speed. This paper focuses on a high level description of the complete system used to solve the overall problem, while providing necessary descriptions of the component algorithms.

Image transform bootstrapping and its applications to semantic scene classification

The performance of an exemplar-based scene classification system depends largely on the size and quality of its set of training exemplars, which can be limited in practice. In addition, in nontrivial data sets, variations in scene content as well as distracting regions may exist in many testing images to prohibit good matches with the exemplars. Various boosting schemes have been proposed in machine learning, focusing on the feature space. We introduce the novel concept of image-transform bootstrapping using transforms in the image space to address such issues. In particular, three major schemes are described for exploiting this concept to augment training, testing, and both. We have successfully applied it to three applications of increasing difficulty: sunset detection, outdoor scene classification, and automatic image orientation detection. It is shown that appropriate transforms and meta-classification methods can be selected to boost performance according to the domain of the problem and the features/classifier used.

Incorporation of derivative priors in adaptive Bayesian color image segmentation

In this study, we attempt to incorporate derivative measures into the scheme of adaptive Bayesian color image segmentation, and provide a unified perspective of region-based and edge-based statistical forces in segmentation. The standard Gibbs random field (GRF) is extended from being determined only by intensity (color) values, i.e., the zero-order derivatives, to also incorporating the first- and second-order derivatives. In particular, a significant first-order derivative accounts for discontinuities to help (1) obtain precise definition of region boundaries, and (2) overcome undesirable under-segmentation caused by the over-smoothing effect often introduced by a standard GRF; an insignificant second-order derivative necessitates stronger homogeneity constraints to help (1) further eliminate small segments caused by noise variations in statistically coherent regions, and (2) overcome undesirable over-segmentation of statistically ramp-like regions. Through balancing the three statistical forces derived from the zero-, first- and second-order derivatives, one can achieve physically and psychophysically more coherent image segmentation.

Normalized Kemeny and Snell distance: a novel metric for quantitative evaluation of rank-order similarity of images

There are needs for evaluating rank order-based similarity between images. Region importance maps from image understanding algorithms or human observer studies are ordered rankings of the pixel locations. We address three problems with Kemeny and Snells distance (d/sub KS/), an existing measure from ordinal ranking theory, when applied to images: its high-computational cost, its bias in favor of images with sparse histograms, and its image-size dependent range of values. We present a novel computationally efficient algorithm for computing d/sub KS/ between two images and we derive a normalized form d/sub KS/ with no bias whose range is independent of image size. For evaluating similarity between images that can be considered as ordered rankings of pixels, d/sub KS/ is subjectively superior to cross correlation.

Displaying images on mobile devices: capabilities, issues, and solutions

Wireless imaging is enabling visual communication ‘anytime anywhere’ to become a reality. Apart from wireless communication issues, a key technical challenge is how to achieve the best-perceived image quality given the limited screen size and display bit-depth of the mobile devices. In this paper, we give an overview of the current capabilities of various mobile devices, highlight some of the technical issues, and present potential solutions. In addition, we present a review of some of the software products on the market and look ahead to the trend toward more capable devices. Copyright

Effect of dielectric layer on the response times of electrostatic MEMS switches

Electrostatic MEMS switches have become prevalent because of low power consumption and ease of integration in micro-fabrication technology. The equations governing their dynamic response obtained by energy methods are nonlinear differential equations. Even the unit-step response of these devices requires numerical computation. Depending on the magnitude of the applied step voltage and the presence of dielectric in the actuator, the response could be recurring or non-recurring. Estimating the period time and the switching time in these cases proves to be hard because one has to solve the energy equation numerically which could be time consuming or difficult to converge if it is not posed properly. Elata et al. have developed excellent methods to obtain these times on a logarithmic scale of voltage more easily for the undamped case. This paper extends their work for the case when the bottom plate is covered with a dielectric layer. The stagnation time occurring before dynamic pull-in, and the switching time thereafter are first shown as nonlinear graphs with the dielectric permittivity as a parameter. They are also linearized on an exponential scale and made useful for quick look up and convenience of designers.

Quantitative evaluation of rank-order similarity of images

Region importance maps from image understanding algorithms and human observer studies are ordered rankings of the pixel locations. Kemeny and Snells distance (d/sub KS/), an existing measure from ordinal ranking theory, can thus be used as a similarity measure between images. We address three problems with d/sub KS/: its high computational cost, its bias in favor of images with sparse histograms, and its image-size dependent range of values. We present a novel computationally efficient algorithm for computing d/sub KS/ between two images, and we derive a normalized form d/sub KS/ with no bias whose range is independent of image size. For evaluating an algorithm where the reference data and algorithm output are ordered rankings of pixels, d/sub KS/ is subjectively superior to the correlation coefficient as a figure of merit.

Relation between charge on free electrodes and the response of electrostatic MEMS actuators and sensors

Stability is an important factor in the study of electrostatic MEMS switches and sensors. Their response is signicantly improved by either applying a large dc bias or by depositing a prescribed value of charge on the oating electrodes. This charge is related to the pull-in voltages. Measurement of charge without causing loading is recommended; so instead of incorporating any eld operated transistor circuitry for this purpose, methods are developed to relate the charge magnitude to the dynamical response of the actuators. Elata et al. developed ecient and reliable ways of charge monitoring without causing loading to the device. These methods rely on energy of the system instead of performing integration in the time domain. Based on their work, this paper examines the alterations in the dynamic response of actuators. The positive and negative pull-in voltages in the voltage displacement plane are symmetrically located with respect to charge on the oating electrode. This fact is exploited to carry out indirect charge measurement from the average of the two pull-in values. A regression scheme is proposed that predicts the charge from the voltage shift based on limited measurements of capacitance of the actuator.

An iterative method for estimating the pull-in parameters of electrostatic actuators

Performance of electrostatic actuators used in MEMS devices is severely limited by the stability considerations that are related to the pull-in parameters. The static and dynamic responses of electrostatic actuators driven by single as well as multiple voltage excitations are studied with an aim of estimating these pull-in voltage and distance parameters. A normalized Hamiltonian formulation is adopted and the resulting equations are solved analytically and also numerically using an iterative scheme. Recently a numerical α-line method has been proposed to extract the pull-in parameters. Scanning along the α-lines by voltage and displacement iteration schemes were studied. Estimating the intersection of the α-lines with the pull-in hypersurface indicates maximal voltage variable. We revisit these two iteration schemes and propose few insights to improve the convergence. Convergence of the parameters to the theoretical values is found to be smooth. This approach helps us to generalize the technique for more complicated geometries.