Micromachined picoliter vials in silicon dioxide with a typical depth of 6.0 /spl mu/m are filled with a liquid sample. Epi-illuminated microscopic imaging during evaporation of the liquid shows dynamic fringe patterns. These fringe patterns are caused by interference between the direct part and the reflected part of an incident plane wave (reflected from the bottom of the vial). The optical path difference (OPD) between the direct and the reflected wave is proportional to the distance to the reflecting bottom of the vial. Evaporation decreases the OPD at the meniscus level and causes alternating constructive and destructive interference of the incident light resulting in an interferogram. Imaging of the space-varying OPD yields a fringe pattern in which the isophotes correspond to isoheight curves of the meniscus. When the bottom is flat, the interference pattern allows monitoring of the liquid meniscus as a function of time during evaporation. On the other hand, when there are objects on the bottom of the vial, the height of these objects are observed as phase jumps in the fringes proportional to their height. First, this paper presents the underlying optical model. Secondly, an image processing method is described to retrieve the meniscus profile from the interference pattern. This algorithm is based on estimating the wrapped (relative) phase of the fringe pattern in the recorded images. Finally, this algorithm is applied to measure height differences on the bottom in other micromachined vials with a precision of about five nanometer.

We present a recursive algorithm for the Gabor filter that achieves, to within a multiplicative constant, the fastest possible implementation. For a signal consisting of N samples, our implementation requires O(N) multiply-and-add operations. Further, the complexity is and coefficients of the recursive equation have a simple independent of the values of /spl sigma/ and /spl omega/ in the Gabor kernel closed-form solution given /spl sigma/ and /spl omega/. Our implementation admits not only a forward Gabor transform from t/spl rarr//spl omega/ but an inverse transform from /spl omega//spl rarr/t that is also O(N) complexity.

We propose a strategy to design recursive implementations of the Gaussian filter and Gaussian regularized derivative filters. Each recursive filter consists of a cascade of two stable Nth-order subsystems (causal and anti-causal). The computational complexity is 2N multiplications per pixel per dimension independent of the size (/spl sigma/) of the Gaussian kernel. The filter coefficients have a closed-form solution as a function of scale (/spl sigma/) and recursion order N (N=3, 4, 5). The recursive filters yield a high accuracy and excellent isotropy in n-D space.

A recent development in fluorescence microscopy is the fluorescence lifetime imaging microscopy. In this type of microscopy the lifetime of the excited state of the fluorescent molecules is of interest rather than the amount of light emitted. By evaluating the response (emission light) of the system (fluorescent molecules) to a specific input signal (excitation light) the fluorescent lifetime can be determined. Different input signals are possible including light pulses, the sinusoidally modulated light and the white-noise modulated light. In this paper we describe two systems under development in our laboratory based on the latter two input signals.

We show that the convergence of measures for object size to their "true values" can be analyzed by looking at the pixels (or voxels) on the border of the object. This leads to results for a proper choice for the sampling density of the asymptotic form CV=/spl sigma//spl mu/=k/spl middot/Q/sup -(N+1/2)/ where Q is the sampling density (pixels per object "diameter") and N is the number of spatial dimensions.

We have compared and improved several implementations of the Wiener filter to remove noise effects from scanning tunneling microscope (STM) images. We have found that the implementation of Weisman et al. (1992), using the noise model of Stoll et al. (1987), provides the best performance on both simulated and real STM images.

We have developed a completely automated fluorescence microscope system that can examine 500 cells in approximately 20 minutes to determine the number of labeled chromosomes (seen as dots) in each cell nucleus. This system works with two fluorescent dyes-one for the DNA hybridization dots (e.g. FITC) and one for the cell nucleus (e.g. DAPI). After the stage has moved to a new field the image is automatically focused, acquired by a Photometrics KAF 1400 camera, and then analyzed on a Macintosh Quadra 840AV computer. After the required number of cells has been analyzed, the user may interact to correct the computer by working with a gallery of the cell images. The machine accuracies are equal to panels of human experts (manual) and limited by the overlapping of dots in the 3D cell as seen through the 2D projection.

In order to investigate the spatio-temporal structure of chromatin in interphase nuclei the authors present two 3-D texture parameters based on the grey-weighted distance transform that quantify the accessibility and the homogeneity of a nucleus. Results of experiments on computer generated textures show that these texture parameters are shape independent.>

In this review of modern image analysis, the author attempts to show how image analysis differs substantially from image processing. He suggests that image analysis can be viewed as a direct measurement or parameter estimation problem or as a problem in image transformation. In the former case great care must be taken in choosing the image representation, estimation formula, and sampling density. In the latter case care must be taken to understand and adapt the transformation to the quantization effects inherent in the digital representation of an image. The relationship between measurement accuracy, the sampling frequency, and the formulas for parameter estimation is studied. It is noted that the levels of accuracy required in industrial inspection, for example, frequently require sampling frequencies far in excess of the Nyquist frequency.>

It is shown that in the context of a specific pattern classification decision metric the number of samplesMneeded to characterize a cluster described byNfeatures isM \geq (1 + \beta^{-1})(N + 2)where\betarepresents an interval width. The distance metricd^{2}(X)=(X-\hat{\mu}_{x})^{t}S_{x}^{-l}(X-\hat{\mu}_{x})is shown to have anF-distribution which leads to the result forM. An additional application of the distribution ofd^{2}(X)is discussed in terms of a specific type of pattern classifier.