Showing papers on "Pixel published in 1994"

Seeded region growing

Abstract: We present here a new algorithm for segmentation of intensity images which is robust, rapid, and free of tuning parameters. The method, however, requires the input of a number of seeds, either individual pixels or regions, which will control the formation of regions into which the image will be segmented. In this correspondence, we present the algorithm, discuss briefly its properties, and suggest two ways in which it can be employed, namely, by using manual seed selection or by automated procedures. >

Hyperspectral image classification and dimensionality reduction: an orthogonal subspace projection approach

Abstract: Most applications of hyperspectral imagery require processing techniques which achieve two fundamental goals: 1) detect and classify the constituent materials for each pixel in the scene; 2) reduce the data volume/dimensionality, without loss of critical information, so that it can be processed efficiently and assimilated by a human analyst. The authors describe a technique which simultaneously reduces the data dimensionality, suppresses undesired or interfering spectral signatures, and detects the presence of a spectral signature of interest. The basic concept is to project each pixel vector onto a subspace which is orthogonal to the undesired signatures. This operation is an optimal interference suppression process in the least squares sense. Once the interfering signatures have been nulled, projecting the residual onto the signature of interest maximizes the signal-to-noise ratio and results in a single component image that represents a classification for the signature of interest. The orthogonal subspace projection (OSP) operator can be extended to k-signatures of interest, thus reducing the dimensionality of k and classifying the hyperspectral image simultaneously. The approach is applicable to both spectrally pure as well as mixed pixels. >

Fast volume rendering using a shear-warp factorization of the viewing transformation

Abstract: Several existing volume rendering algorithms operate by factoring the viewing transformation into a 3D shear parallel to the data slices, a projection to form an intermediate but distorted image, and a 2D warp to form an undistorted final image. We extend this class of algorithms in three ways. First, we describe a new object-order rendering algorithm based on the factorization that is significantly faster than published algorithms with minimal loss of image quality. Shear-warp factorizations have the property that rows of voxels in the volume are aligned with rows of pixels in the intermediate image. We use this fact to construct a scanline-based algorithm that traverses the volume and the intermediate image in synchrony, taking advantage of the spatial coherence present in both. We use spatial data structures based on run-length encoding for both the volume and the intermediate image. Our implementation running on an SGI Indigo workstation renders a 2563 voxel medical data set in one second. Our second extension is a shear-warp factorization for perspective viewing transformations, and we show how our rendering algorithm can support this extension. Third, we introduce a data structure for encoding spatial coherence in unclassified volumes (i.e. scalar fields with no precomputed opacity). When combined with our shear-warp rendering algorithm this data structure allows us to classify and render a 2563 voxel volume in three seconds. The method extends to support mixed volumes and geometry and is parallelizable.

Radiometric CCD camera calibration and noise estimation

Abstract: Changes in measured image irradiance have many physical causes and are the primary cue for several visual processes, such as edge detection and shape from shading. Using physical models for charged-coupled device (CCD) video cameras and material reflectance, we quantify the variation in digitized pixel values that is due to sensor noise and scene variation. This analysis forms the basis of algorithms for camera characterization and calibration and for scene description. Specifically, algorithms are developed for estimating the parameters of camera noise and for calibrating a camera to remove the effects of fixed pattern nonuniformity and spatial variation in dark current. While these techniques have many potential uses, we describe in particular how they can be used to estimate a measure of scene variation. This measure is independent of image irradiance and can be used to identify a surface from a single sensor band over a range of situations. Experimental results confirm that the models presented in this paper are useful for modeling the different sources of variation in real images obtained from video cameras. >

Sub-pixel land cover composition estimation using a linear mixture model and fuzzy membership functions

Abstract: Mixed pixels occur commonly in remotely-sensed imagery, especially those with a coarse spatial resolution. They are a problem in land-cover mapping applications since image classification routines assume ‘pure’ or homogeneous pixels. By unmixing a pixel into its component parts it is possible to enableinter alia more accurate estimation of the areal extent of different land cover classes. In this paper two approaches to estimating sub-pixel land cover composition are investigated. One is a linear mixture model the other is a regression model based on fuzzy membership functions. For both approaches significant correlation coefficients, all >0·7, between the actual and predicted proportion of a land cover type within a pixel were obtained. Additionally a case study is presented in which the accuracy of the estimation of tropical forest extent is increased significantly through the use of sub-pixel estimates of land-cover composition rather than a conventional image classification.

Cmos image sensor with pixel level a/d conversion

Abstract: An image sensor formed using a CMOS process is described herein which includes a pixel array core of phototransistors whose conductivities are related to the magnitude of light impinging upon the phototransistors. The analog signals generated by the phototransistors are converted to a serial bit stream by an A/D converter connected at the output of each phototransistor and formed in the immediate area of each phototransistor within the array core. Thus, a separate digital stream for each pixel element is output from the array core, and parasitic effects and distortion are minimized. In one embodiment, a filter circuit is connected to an output of the array core for converting the individual digital streams from each pixel element to multi-bit values corresponding to the intensity of light impinging on the phototransistor.

Original approach for the localisation of objects in images

Abstract: An original approach is presented for the localisation of objects in an image which approach is neuronal and has two steps. In the first step, a rough localisation is performed by presenting each pixel with its neighbourhood to a neural net which is able to indicate whether this pixel and its neighbourhood are the image of the search object. This first filter does not discriminate for position. From its result, areas which might contain an image of the object can be selected. In the second step, these areas are presented to another neural net which can determine the exact position of the object in each area. This algorithm is applied to the problem of localising faces in images.

Multi-color information encoding system

Abstract: An exceptionally dense information encoding system, with 4-10 times the density of CD ROM diskettes, employs colored areas in the form of bars or checkerboard matrices of colored dot regions to encode information including alphanumerics, with each colored region being variable as to both color and intensity. In one embodiment, so-called super pixel dots have differently colored sub-regions within them, arranged with side-by-side colors or with colored regions stacked one on top of the other, such that information from one dot has as many color variables as there are stacked layers or mixed colors. In one embodiment the super pixel dot is 5 microns in diameter with 2 micron spacing between adjacent dots. For each color in one embodiment there are as many as 64 intensities yielding a coding system of high information density. The various colors are read out at one super pixel dot position by dividing out reflected or transmitted energy from a dot by color filtering such that a color and intensity can be detected for each color intensity within the super pixel dot. The code provided by the subject system is substantially invisible to the naked eye, with machine vision and computer analysis of the information being required to effectively decode differences of intensity. Additionally, encrypting can be performed on the digitally encoded information to further hide the information carried by the colored dot matrix. Moreover, standardized intensities are established by one or more of the coded regions to assist in machine decoding, and correction techniques are applied for variations in detected color and intensity.

Accelerating Volume Reconstruction With 3D Texture Hardware

Abstract: This paper describes a new method of rendering volumes that leverages the 3D texturing hardware in Silicon Graphics RealityEngine workstations The method defines the volume data as a 3D texture and utilizes the parallel texturing hardware to perform reconstruction and resampling on polygons embedded in the texture The resampled data on each polygon is transformed into color and opacity values and composited into the frame buffer A 128×128×64 volume is rendered into a 512 window at over 10 frames per-second Two alternative strategies for embedding the resampling polygons are described and their trade-offs are discussed This method is easy to implement and we apply it to the production of digitally reconstructed radiographs as well as opacity-based volume rendered images The generality of this approach is demonstrated by describing its application to the proposed PixelFlow graphics system PixelFlow overcomes the lighting and volume size limitations imposed by the RealityEngine It is expected to render 256 data sets on a 640×512 screen at over 10 frames per second 1 Fig 1 Resampling polygon orientations Volume Boundary a) Object space sample planes b) Image space sample planes RealityEngine is a trademark of Silicon Graphics Inc where ui are the resample values behind a pixel and d is the spacing between sample values Note that d is constant for all samples behind a pixel, but due to perspective, it varies from pixel to pixel The resampled ui terms are summed at each pixel, and the d factors are applied by using an additional full-screen polygon with a 2D texture corresponding to the d values required for each pixel The summation results may be viewed directly or the exponential required to mimic a radiograph may be computed at each pixel by using a lookup table The RealityEngine has a maximum precision of 12bits per frame buffer and texture component The summation could easily overflow that unless the sample values are properly scaled Our implementation maintains 12-bit volume data values in the texture memory and scales each resampled value by a user controlled "exposure" value ranging from zero to one The scaled samples are then summed and clamped if they exceed the 12-bit range In practice, it has been easy to find suitable exposure control settings for the data sets tested Figure 2 shows radiograph images of 128×128×64 CT data of a human pelvis made with polygons aligned in object-space polygons aligned with the object-space axes Figure 1b shows resampling on polygons aligned with the image-space axes In either case, the resampled values behind each pixel are combined to produce a color for that pixel The combining method is often a compositing operation, but may be other operations as required by the visualization application Polygons aligned in object-space are defined to lie within the volume and rendered with GL library calls This method is complicated slightly by the need to reorient the sampling polygons in the plane most parallel to the view-plane as the view-point changes This is accomplished by examining the view matrix and explicitly creating polygons for the six cases that arise [Westover91] Polygons aligned in image-space must be clipped to the boundaries of the volume to ensure valid texture coordinates Polygons are defined in image-space and transformed by the inverse viewing matrix into objectspace where the clipping occurs Clipped polygons are then rendered with the usual GL library calls In addition to using unclipped polygons, there are other advantages to using the object-space method The texturing time for any polygon is proportional to the number of pixels it covers A priori information about the extent of interesting features in each slice of the volume may be used to minimize the polygon size, and thus its texturing time, as a function of its location The texture memory of the RealityEngine is limited to 1M 12-bit data points To render larger volumes, slab subsets are loaded and rendered in succession Texture memory may be reloaded in about 01 seconds With the object-space method, rendering each slab is simple The image-space method must render polygons multiple times, clipping them to the currently loaded volume slab 3 Radiographs A digitally reconstructed radiograph of medical volume data is produced by combining the resampled values behind each pixel to approximate the attenuation integral pixel intensity = 10 exp(-∑ ui d ) (1) 2 Fig 3 Digitally reconstructed radiographs is not normalized and normalization is an expensive process requiring a square root Lighting without normalization is possible, but this has not yet been tried to see how serious the artifacts are 5 Performance We consider two data sizes rendered into a 512 window The smaller data size of 128×128×64 may be rendered at ten frames per-second using 128 polygons aligned in object-space This equates to a processing rate of 10 million voxels per-second In our test images we measured about 160 million pixel operations per second, where each pixel operation is a trilinear interpolation of the 3D texture components, a multiplication by a scaling or opacity factor, and a summation or composite into the frame buffer The larger data size of 256×256×64 requires four 256×256×16 texture slabs and is rendered at 25 frames per-second with 256 resampling polygons Loading texture slabs consumes less than 01 seconds per-slab A progressive refinement approach would allow a user to manipulate the low-resolution data at a high frame rate, and render the high-resolution data as soon as the user allows the motion to stop The performance is very linear with respect to the number of pixels processed As the number of screen pixels or resampling polygons is doubled, the frame rate is halved If more resampling polygons are used, higher quality images are obtained at the expense of lower rendering speed 6 PixelFlow Texturing hardware is likely to be a common feature of graphics systems in the future The PixelFlow graphics system under development at the University of North Carolina at Chapel Hill will have texturing hardware [Molnar92] that it is suitable for a variant of the polygon resampling approach described above for the RealityEngine We propose a polygon texturing approach for the PixelFlow system that will overcome the limitations on realistic lighting and data size imposed by the RealityEngine The texturing hardware in PixelFlow will allow 128 pixel processors to access eight arbitrarily-addressed 32-bit values in texture memory in under 500 μs PixelFlow texturing hardware does not perform any operations on these texture values; rather, they are simply loaded into the pixel processors where a user’s program manipulates them as ordinary data If the 32-bit values are treated as four 8-bit texture components, then three may be 4 Opacity-Based Rendering The summation of samples produces radiograph images Compositing samples produces images with occlusion Only one texture component is required for the linear attenuation coefficient used to produce radiographs Two 8-bit texture components can represent the raw data and a precomputed shading coefficient The resampled data component values are used as indices into an opacity lookup table This lookup uses the texture hardware for speed The shading coefficient is a function of the original data gradient and multiplies the sample opacity to produce images of shaded features as shown in figure 3 This figure shows the human pelvis data set above an image of a 4 mm volume of a chicken embryo acquired by a microscopic MRI scanner The precomputed shading fixes the light position(s) relative to the volume For more general lighting by sources fixed in image-space, the shade texture component must be replaced by three components containing the normalized data gradient Unfortunately, the resampled gradient on the polygons Fig 3 Shaded volume rendering

Pixel control circuitry for spatial light modulator

Abstract: A spatial light modulator (10) having reduced control circuitry as compared to existing devices. Sets of pixel elements (11) share a memory cell (12), such that each memory cell (12) has the same fanout as other memory cells (12). Each pixel element (11) in a set is switched to an on or off state via a reset line (13) that is separate from that of the other pixel elements (11) in that set. Frame data is loaded in split bit-frames during a set time period, such that each split bit-frame contains only the data for pixel elements (11) on one reset line (13). Thus, the same memory cell (12) can be used to deliver data to all pixel elements (11) in its fanout because only one pixel element (11) in the fanout is switched at a time.

Fast segmentation of range images into planar regions by scan line grouping

Abstract: A novel technique is presented for rapid partitioning of surfaces in range images into planar patches. The method extends and improves Pavlidis' algorithm (1976), proposed for segmenting images from electron microscopes. The new method is based on region growing where the segmentation primitives are scan line grouping features instead of individual pixels. We use a noise variance estimation to automatically set thresholds so that the algorithm can adapt to the noise conditions of different range images. The proposed algorithm has been tested on real range images acquired by two different range sensors. Experimental results show that the proposed algorithm is fast and robust.

Eye tracking apparatus and method employing grayscale threshold values

Abstract: An eye-tracking system determines the position of a user's pupil and maps this position into a point of regard of the user on an interface device, such as a display screen, or other real-world object by a system comprising a camera for acquiring a video image of the pupil; a frame grabber coupled to the camera for accepting and converting analog video data from the camera to digital pixel data; a computer coupled to the frame grabber for processing the digital pixel data to substantially determine the position of the pupil; a display screen coupled to the computer; and a support connected to the camera and display screen for fixing the relative physical positions thereof relative to the user's pupil. The processing performed by the computer may include the selection of a first pixel intensity threshold for the segmentation of the digital pixel data into first and second groups, where the total pixel area of the first group is selected to be substantially equal to a pre-determined value expected to correspond to the area of a user's pupil. The system may be calibrated by the user's following a cursor on the display screen while the system measures the pupil position for known locations of the cursor.

System for encoding image data into multiple layers representing regions of coherent motion and associated motion parameters

Abstract: A system stores images as a series of layers by determining (i) the boundaries of regions of coherent motion over the entire image, or frame, sequence; and (ii) associated motion parameters, or coefficients of motion equations, that describe the transformations of the regions from frame to frame. The system first estimates motion locally, by determining the movements within small neighborhoods of pixels from one image frame i to the next image frame i+1, to develop an optical flow, or dense motion, model of the image. Next, the system estimates the motion using affine or other low order, smooth transformations within a set of regions which the system has previously identified as having coherent motion, i.e., identified by analyzing the motions in the frames i-1 and i. It groups, or clusters, similar motion models and iteratively produces an updated set of models for the image. The system then uses the local motion estimates to associate individual pixels in the image with the motion model that most closely resembles the pixel's movement, to update the regions of coherent motion. Using these updated regions, the system iteratively updates its motion models and, as appropriate, further updates the coherent motion regions, and so forth. The system then does the same analysis for the remaining frames. The system next segments the image into regions of coherent motion and defines associated layers in terms of (i) pixel intensity values, (ii) associated motion model parameters, and (iii) order in "depth" within the image.

Method and apparatus for machine vision classification and tracking

Abstract: A method and apparatus for classification and tracking objects in three-dimensional space is described. A machine vision system acquires images with a video camera (2) from roadway scenes (6) and processes the images by analyzing the intensities of edge elements within the image. The system then applies fuzzy set theory to the location and angles of each pixel after the pixel intensities have been characterized by vectors. A neural network interprets the data created by the fuzzy set operators and classifies objects within the roadway scene (6). The system also includes a tracking module (22) for tracking objects within the roadway scene, such as vehicle, by forecasting potential track regions and then calculating match scores for each potential track region based on how well the edge elements from the target track regions match those from the source region as weighted by the extent the edge elements have moved.

Electronic speckle photography: increased accuracy by nonintegral pixel shifting.

Abstract: Electronic speckle photography offers a simple and fast technique for measuring in-plane displacement fields in solid and fluid mechanics. An improved algorithm is presented and analyzed by use of both computer-simulated speckle patterns and real experiments. The idea of the improved algorithm is to maximize the correlation between correlated subimages from different images by shifting one of them by nonintegral pixel values. The improved algorithm was found to determine displacement components with an uncertainty of less than 1% of a pixel and with negligible systematic errors in ideal experimental conditions.

Digital micromirror device and its application to projection displays

Abstract: The digital micromirror device (DMD) is a monolithic, micromechanical spatial light modulator. The DMD has been used to implement the first truly digital projection display systems. In these systems, discrete, tilting mirror elements are fabricated from sputter deposited aluminum directly on top of arrays of complementary metal–oxide semiconductor memory cells. The mirrors are switched between two stable tilted states according to whether a ‘‘1’’ or a ‘‘0’’ is stored in the underlying memory location. An optical system illuminates the DMD and projects its image in such a way that the image of each mirror, which represents a single pixel in the projected image, is at full brightness when the mirror is tilted in the ‘‘1’’ state and full darkness when the mirror is tilted in the ‘‘0’’ state. The refresh rate of the memory and the response rate of the mirrors are high enough so that hundreds of memory frames can be displayed during one video frame, and so that each pixel can be on or off in a binary fashion for a portion of the video frame proportional to that pixel’s individual intensity. The digital‐to‐analog conversion of this intensity occurs in the eye/brain of the viewer. The mirrors are typically square, 16 μm on a side, and placed on 17 μm centers. Each mirror tilts 10° from horizontal in each of its two addressed states, so that the ‘‘1’’ state and ‘‘0’’ state are 20° apart. Arrays of mirrors ranging from resolutions of 768×576 mirrors up to 2048×1152 mirrors have been fabricated.The article will describe the fabrication process for the DMD, the optical system used to project the DMD image, and the electronic method of addressing the device. Prototype projection systems will be described and preliminary performance measurements will be presented.

Video scaling method and device

Abstract: A linear interpolator and decimating FIR filter with constant coefficients may be used in various configurations to scale a still or moving color video image horizontally (changing the number of pixels per row) or vertically (changing the number of pixels per column). For downscaling, the input pixel stream is initially upscaled with an interpolator by an upscaling factor of between one and two chosen so that the result can be subsequently downscaled with a decimation-by-power-of-two finite impulse response (FIR) filter to provide the desired final scaling. Upscaling is a special case accomplished by means of linear interpolation. Vertical and horizontal scaling may be performed with different, independent scaling factors.

Dmd architecture to improve horizontal resolution

Abstract: A method and device for increasing the effective horizontal resolution of a display device. One embodiment of the invention forms a cardinal array of digital micromirror elements by staggering alternate rows in an array. According to a second embodiment of the claimed invention, an ordinal pixel array, is converted to a cardinal pixel array, by grouping SLM elements into a pixel block. All of the elements in a pixel block are controlled in unison such that the pixel block acts like a single pixel. Rows of pixel blocks axe offset to provide the effect of a cardinal array of pixels without the decrease in efficiency sometimes associated with cardinal pixel arrays.

High resolution, high brightness light emitting diode display and method and producing the same

Abstract: A high resolution, high brightness, full color display is provided having a liquid crystal pixel selectably addressable during a predetermined time period, a set of at least one red, one green, and one blue color light emitting diodes positioned adjacent the liquid crystal pixel for emitting light through the liquid crystal pixel, and means connected to the liquid crystal pixel for addressing the liquid crystal pixel a plurality of times during the predetermined time period for each color so as to provide persistence when changes in color are perceived by the human eye. A method of producing a high resolution, full color display is also provided by lighting a set of one red, one green, and one blue light emitting diodes by lighting the respective colored light sources for a predetermined time period for each color and shuttering the set of light sources with a liquid crystal pixel for at least a portion of the predetermined time period to thereby emit light from the shuttered pixel for a selected time period so as to provide persistence when changes in color are perceived by the human eye.

A CMOS area image sensor with pixel-level A/D conversion

Abstract: Charge-coupled devices (CCD) are at present the most widely used technology for implementing area image sensors. However, they suffer from low yields, consume too much power, and are plagued with SNR limitations due to the shifting and detection of analog charge packets, and the fact that data is communicated off chip in analog form. This paper describes an area image sensor that can potentially circumvent the limitations of CCDs and their alternatives. It uses a standard CMOS process and can therefore be manufactured with high yield. Digital circuitry for control and signal processing can be integrated with the sensor. Moreover, CMOS technology advances such as scaling and extra layers of metal can be used to improve pixel density and sensor performance. The analog image data is immediately converted to digital at each pixel using a one-bit sigma-delta modulator. The use of sigma-delta modulation allows the data-conversion circuitry to be simple and insensitive to process variations. A global shutter provides variable light input attenuation to achieve wide dynamic range. Data is communicated off chip in a digital form, eliminating the SNR degradation of analog data communication. To demonstrate the viability of the approach, an area image sensor chip is fabricated in a 1.2 /spl mu/m CMOS technology. The device consists of an array of 64x64 pixel blocks, a clock driver, a 6:64 row address decoder, 64 latched sense amplifiers, and 16 4:1 column multiplexers. The chip also contains data compression circuitry. >

Film-to-video format detection for digital television

Abstract: A film-to-video format detector (24) for a digital television receiver (10). The detector (24) receives pixel data from a current field and a second preceding field. It determines a set of pixel difference values, sums them to obtain a field difference value, and compares the field difference value to a threshold. These steps are repeated to obtain a series of field difference indicators. This series is analyzed to determine whether it has a pattern corresponding to a film-to-video format.

Frameless rendering: double buffering considered harmful

Abstract: The use of double-buffered displays, in which the previous image is displayed until the next image is complete, can impair the interactivity of systems that require tight coupling between the human user and the computer. We are experimenting with an alternate rendering strategy that computes each pixel based on the most recent input (i.e., view and object positions) and immediately updates the pixel on the display. We avoid the image tearing normally associated with single-buffered displays by randomizing the order in which pixels are updated. The resulting image sequences give the impression of moving continuously, with a rough approximation of motion blur, rather than jerking between discrete positions.We have demonstrated the effectiveness of this frameless rendering method with a simulation that shows conventional double-buffering side-by-side with frameless rendering. Both methods are allowed the same computation budget, but the double-buffered display only updates after all pixels are computed while the frameless rendering display updates pixels as they are computed. The frameless rendering display exhibits fluid motion while the double-buffered display jumps from frame to frame. The randomized sampling inherent in frameless rendering means that we cannot take advantage of image and object coherence properties that are important to current polygon renderers, but for renderers based on tracing independent rays the added cost is small.

Video test strip reader and method for evaluating test strips

Abstract: A video test strip reader uses a video imager or camera for viewing a viewing field containing reagent test strips each having test pads reacted with a specimen containing constituents of interest. The video imager produces an analog signal representing an image of the viewing field. An image handler coupled to the video imager converts or digitizes the analog signal into a digital signal representing the image and stores the image in the form of an array of pixels representing the image. Each pixel contains color information broken down into red, green or blue (RGB). A processor coupled to the image handler analyzes the array of pixels, determines the location and orientation of a test strip, identifies the test areas on the test strip, measures the corresponding test areas on the strip at the proper times and calculates the test results, such as the concentration of the constituents of interest in the specimen or other measurable properties of the specimen such as color or specific gravity, etc. Accordingly, the video test strip reader can simultaneously locate, color analyze and time track multiple test strips on the viewing field.

Color image gamut-mapping system with chroma enhancement at human-insensitive spatial frequencies

Abstract: A method and system for gamut-mapping color images from device-independent form to device-dependent gamut in a Cartesian color space. Digital images are mapped to any device-dependent gamut in a manner that minimizes the human visual response both to the luminance and the chrominance changes necessary to force out-of-gamut pixels into the specified device-dependent gamut. A "neighborhood gamut mapping" technique considers the subjective visual effects of nearby pixels on the mapping of each pixel. Image luminance is biased toward the luminance in the device-dependent gamut at which the greatest chromamagnitude is available for a fixed hue angle. The chrominance of the mapped image is thereby imperceptibly adjusted to compensate for the human visual effects of luminance changes. Spatial filtering exploits the differing spatial frequency regions of insensitive human visual response to both luminance and chrominance changes.

Hierarchical spline-based image registration

Abstract: The problem of image registration subsumes a number of topics in multiframe image analysis, including the computation of optic flow (general pixel-based motion), stereo correspondence, structure from motion, and feature tracking. We present a new registration algorithm based on a spline representation of the displacement field which can be specialized to solve all of the above mentioned problems. In particular, we show how to compute local flow, global (parametric) flow, rigid flow resulting from camera egomotion, and multiframe versions of the above problems. Using a spline-based description of the flow removes the need for overlapping correlation windows, and produces an explicit measure of the correlation between adjacent flow estimates. We demonstrate our algorithm on multiframe image registration and the recovery of 3D projective scene geometry. We also provide results on a number of standard motion sequences. >

Edge-region-based segmentation of range images

Abstract: In this correspondence, we present a new computationally efficient three-dimensional (3-D) object segmentation technique. The technique is based on the detection of edges in the image. The edges can be classified as belonging to one of the three categories: fold edges, semistep edges (defined here), and secondary edges. The 3-D image is sliced to create equidepth contours (EDCs). Three types of critical points are extracted from the EDCs. A subset of the edge pixels is extracted first using these critical points. The edges are grown from these pixels through the application of some masks proposed in this correspondence. The constraints of the masks can be adjusted depending on the noise present in the image. The total computational effort is small since the masks are applied only over a small neighborhood of critical points (edge regions). Furthermore, the algorithm can be implemented in parallel, as edge growing from different regions can be carried out independently of each other. >

Efficient maximum likelihood classification for imaging spectrometer data sets

Abstract: A simplified maximum likelihood classification technique for handling remotely sensed image data is proposed which reduces, significantly, the processing time associated with traditional maximum likelihood classification when applied to imaging spectrometer data, and copes with the training of geographically small classes. Several wavelength subgroups are formed from the complete set of spectral bands in the data, based on properties of the global correlation among the bands. Discriminant values are computed for each subgroup separately and the sum of discriminants is used for pixel labeling. Several subgrouping methods are investigated and the results show that a compromise among classification accuracy, processing time, and available training pixels can be achieved by using appropriate subgroup sizes. >

Image data compression having minimum perceptual error

Abstract: A method for performing image compression that eliminates redundant and invisible image components. The image compression uses a Discrete Cosine Transform (DCT) and each DCT coefficient yielded by the transform is quantized by an entry in a quantization matrix which determines the perceived image quality and the bit rate of the image being compressed. The present invention adapts or customizes the quantization matrix to the image being compressed. The quantization matrix comprises visual masking by luminance and contrast techniques and by an error pooling technique all resulting in a minimum perceptual error for any given bit rate, or minimum bit rate for a given perceptual error.

Enhanced resolution for digital micro-mirror displays

Abstract: A display system (20) that uses multiple SLMs (25) to enhance horizontal or vertical resolution, or both. For example, to approximate a two-fold increase in horizontal resolution, the input data is sampled at a doubled rate, and each SLM (25) receives every other sample. Each SLM (25) generates an image, and the two images are partially superposed with a horizontal offset and simultaneously displayed. The resulting output image has a perceived resolution that approximates that of an image generated by an SLM with twice as many pixels per row.