Single chip charge-coupled devices (CCDs) coupled with filters for isolating red, green, and blue color content are commonly used to capture color images. While this is more cost effective than multiple chip systems, best results are obtained when full RGB color information is obtained for every point in an image. The process of color subsampling in a single chip system degrades the resulting image data by introducing artifacts such as blurry edges and false coloring. We propose an algorithm for enhancing color image data that were captured with a typical single chip CCD array. The algorithm is based on stochastic regularization using a Markov random field model for the image data. This results in a constrained optimization problem, which is solved using an iterative constrained gradient descent computational algorithm. Results of the proposed algorithm show a marked improvement over the original sampled image data.

We present a new approach for improving the effective dynamic range of cameras by using multiple photographs of the same scene taken with different exposure times. Using this method enables the photographer to accurately capture scenes that contain high dynamic range by using a device with low dynamic range, which allows the capture of scenes that have both very bright and very dark regions. We approach the problem from a probabilistic standpoint, distinguishing it from the other methods reported in the literature on photographic dynamic range improvement. A new method is proposed for determining the camera's response function, which is an iterative procedure that need be done only once for a particular camera. With the response function known, high dynamic range images can be easily constructed by a weighted average of the input images. The particular form of weighting is controlled by the probabilistic formulation of the problem, and results in higher weight being assigned to pixels taken at longer exposure times. The advantages of this new weighting scheme are explained by comparison with other methods in the literature. Experimental results are presented to demonstrate the utility of the algorithm.

We describe a multiresolution, dynamic, and adaptive image quantization methodology with automation being the goal of our research. To improve the robustness of the approach, we incorporate dynamic local thresholding and multiresolution peak detection. The first strategy extracts bisector values from local regions of the image and builds a histogram based on those values. The second strategy maps the derived histogram into multiple levels of resolution, allowing peaks be scored for their significance and localized. We conduct several experiments to analyze different versions of our quantization methodology and to compare it with the equal probability quantization. We also investigated the relationships between image attributes and the key parameters in our quantizers. Based on the findings, we developed a fully automated quantizer called QTR0.5. We have applied QTR0.5 to a variety of images—aerial, photographic, and satellite images—and have also used it as a preprocessor in an image segmentation software tool.

We present super-resolution methods that enhance image contrast and perform anisotropic diffusion simultaneously. Since we solve the super-resolution problem by encouraging Mach-band profiles while incorporating anisotropic diffusion, our technique not only reconstructs a high-resolution image from several overlapping noisy low-resolution images, but also enhances edges and image contrast while suppressing image noise during the restoration process. We apply this technique to a video stream, which can be aligned by 3×3 projective transformations.

Although it would seem obvious that object-oriented techniques are well suited to modeling image processing systems, there is almost no research to substantiate this position. We review the unique challenges faced when modeling, that is, specifying and designing, imaging systems. First, the argument is presented that specification and design of imaging systems should be studied as a special case. Next, a subset of techniques that have been used to model practical imaging systems is surveyed. Then the use of object-oriented techniques in the specification of imaging systems is investigated more fully and compared to a widely used alternative—structured analysis and design—by way of a case study. We conclude with some recommendations for best practices and future research.

Median filtering has proved an effective way to remove impulse noise while preserving rapid signal changes. A classic general purpose median filter is based on a bubble sort approach. We present a novel bit-level algorithm that is readily scalable and very compact. The basis of the algorithm is similar to that of Quicksort, and is based on a bit voter (BV) block. The work also extends the basic BV algorithm to include weighted median and ranked order median filtering. The median finding unit, whose inputs are the N pixel values to sort, has an O(N) hardware complexity compared to an O(N²) complexity for an equivalent unit using bubble sort. The novel algorithm has been implemented on a Xilinx Spartan XCS30XL-4 FPGA chip. For comparison purposes, we also present an FPGA implementation of an existing triple input sorter based algorithm (TIS), which is an optimized version of the bubble sort algorithm for the special case of a 3×3 window size. The BV median finding unit occupies 15 configurable logic blocks (CLBs) only, whereas TIS occupies 60 CLBs. For PAL video (720×576 images of 8-bit/pixel), comparative timings show that BV can operate at 26 frames/sec (fps), while TIS can achieve 33 fps. Thus, both algorithms can achieve real-time performance, and BV is the more compact.

We present a low-power 3-D discrete wavelet transform processor for medical applications. The main focus is the compression of medical resonance image (MRI) data, although the system could be used as a generic compression chip. The architecture eliminates redundant filter banks by using a central control unit to dynamically adjust filter parameters. An on-chip cache is used to process block inputs minimizing result throughput. Power consumption has been kept to a minimum by placing constraints throughout the entire design process. The modular processor has been prototyped using 0.6-μm complementary metal oxide semiconductor (CMOS) (three metal) technology. It has been simulated at the functional, circuit, and physical levels. The performance measures of the prototype, area, time delay, power, and utilization have been evaluated. The prototype operates at an estimated frequency of 272 MHz and dissipates 0.5 W of power.

We present some novel tools for the analysis of blue-noise binary patterns. Unlike most of the existing methods that evaluate the frequency content of a given mask or its lower order statistics, our new metrics characterize the morphological content of a mask that is quantified using simple one-pass filtering. An analytical filter expression is given. As a result, one can balance the structural content of the mask—diagonal, vertical, and horizontal interconnections of the majority (or minority) pixels—at the same level. In addition, it is possible to improve the overall mask quality by prescribing the occurrence of morphological shapes of connected pixels. Examples of morphological analysis are given to demonstrate the different qualities of blue-noise and white-noise patterns.

It is recognized that block-wise dependence is a key requirement for fragile watermarking schemes to thwart vector quantization attack. It has also been proved that dependence with deterministic or limited context is susceptible to transplantation attack or even simple cover-up attacks. In this work, we point out that traditional nondeterministic block-wise dependence is still vulnerable to cropping attacks and propose a 1-D neighborhood forming strategy to tackle the problem. The proposed strategy is then implemented in our new fragile watermarking scheme, which does not resort to cryptography and requires no a priori knowledge about the image for verification. To watermark the underlying image, the gray scale of each pixel is adjusted by an imperceptible quantity according to the consistency between a key-dependent binary watermark bit and the parity of a bit stream converted from the gray scales of a secrete neighborhood formed with the 1-D strategy. The watermark extraction process works exactly the same as the embedding process, and produces a difference map as output, indicating the authenticity and integrity of the image.

We present an adaptive method that allows the classification of the regions of an image in two classes: textured and uniform regions (weak gray-level variance). Information on the type of texture (stochastic, deterministic) and its granularity (macroscopic, microscopic) are extracted. The developed method is achieved in two steps. It first enables the determination of the global context of an image (image mainly composed of uniform areas or textured ones) and the localization of the textured and uniform areas. The second step characterizes each detected area by considering some appropriate features: mean and variance in the case of uniform regions and classical relevant texture attributes (derived from a statistical analysis) associated with some new attributes that we define in the case of textured regions. These complementary features are determined from a texture model derived from the Wold decomposition of the autocovariance function. They enable the acquisition of some information on the type of texture and its granularity. We show the efficiency of the method through two examples of image segmentation.

Inspired by the idea that threshold surface always intersects the image surface at high gradient points, an active surface-based adaptive thresholding algorithm is proposed to get the binarized result. In this model, the external force is designed to be repulsive from the image surface, thus at the equilibrium state the active surface tends to cover the supporting points of high gradient with smooth property, as well as be away from the image surface locally, which makes the obtained threshold surface properly separate the foreground and background. The description of the algorithm is in a simple and reasonable energy functional form, and only two parameters need to be tuned, which gives more convenience to the operation. Analysis and comparison for the experimental results reveal that it cannot only give the proper thresholding result but also restrain the occurrence of the ghost phenomenon.

A new anisotropic diffusion approach is proposed to segment texture images. The approach uses an annealing function as a parameter for the conductance coefficient in an anisotropic diffusion to smooth edges within texture regions and preserve boundaries between regions. The smoothing behavior of the diffusion using different annealing functions is analyzed in detail. The proposed method is first applied to segment texture images based on only image intensities. Then, by extending it to a vectorized diffusion to operate on a texture feature vector, more complex texture images can be segmented.

Leg ulcers are chronic skin wounds that affect many people and take a long time to heal. The progress of wound healing and the effect of clinical treatments can be monitored partly by measuring the area of the wound. Measurements taken via manually based methods, such as using a computer pointing device to delineate the wound boundary in a digitized image, suffer from variations due to manual dexterity and differences of opinion between observers. Two active contour models are presented, and their measurement performances are compared with that of the existing manual delineation method. The models make use of manual delineation to initialize the solution and are shown to reduce the effect of the inherent variations on the repeatability (precision) of area measurements in most cases. Measurements made by both algorithms are often biased with respect to the manual measurements, although the bias is limited generally to less than 5% of the wound's area.

A method for understanding convex objects is presented. Convex objects frequently appear in many fields of research and engineering. Research into the problem of understanding convex objects has been carried out as part of wider investigations related to the method of shape understanding, implemented as a shape understanding system (SUS). The main novelty of the presented method is that the process of understanding a convex object is related to the visual concept represented as a symbolic name of the possible classes of shapes. The possible classes of shapes are viewed as a hierarchical structure, where at each level of description the different aspects of shape are incorporated into the shape model. The new "intelligent" way of processing is proposed, in which at each stage of the reasoning process that leads to assigning an examined object to one of the possible classes, the specialized processing methods are used. These methods are very efficient because they deal with a specific class of objects. The proposed shape understanding system is designed to interpret visual data in terms of shape properties to perform tasks that require higher levels of understanding capabilities.

We explore a cost-effective approach to creating environment maps. Unlike other environment map generating systems that use specially designed mirrors or sensing devices, our system uses off-the-shelf digital cameras. This makes our system potentially acquirable by average consumers. In our design, the two digital cameras are displaced vertically with parallel optical axes, and the rotation axis passes through the focal points of the two cameras. This vertical displacement arrangement overcomes the problem of having inconsistent occlusions in two adjacent stereo image sets that happen in a horizontally displaced stereo camera system. To obtain dense spatial images, our system uses color-information-based algorithms to propagate disparity values from matched feature points to unmatched feature points and featureless areas. Schemes of transforming local 3-D (X,Y,Z) data to a world coordinate system are also devised to smoothly stitch 3-D data from individual frames for virtual reality modeling language graphics display.

The World Wide Web is incorporating recent advances in Internet technology and 3-D graphics, promoting the spread of visualization over networks. To increase the efficiency of web-based 3-D graphics systems including scientific visualizations, we present a method that utilizes previously rendered and transmitted images instead of transmitting 3-D models to the client. Our image-based rendering and compression (IBRAC) method exploits spatial and temporal coherence between new and previously rendered synthetic images. Reprojecting color and visibility data accelerates the computation of new images, while reprojecting the surface orientation eliminates inconsistencies in shading and lighting changes that plague other image-based rendering methods. We also present a new image data structure called an isomap, containing classification-interval data that enables a remote user to interactively classify volume data without loading the 3-D dataset. Our approach produces rendering speedups of seven and more without visible degradation, and compression ratios are a factor of 2 to 10 times better than MPEG2 in our test cases. The approach is best suited for remote rendering applications, where the client has limited rendering resources and network bandwidth, including wireless or palm computing applications.