Grayscale Imagery Research Articles

Development of morphological operators for field programmable gate arrays

In this paper, we'll investigate development of Morphological Operators (MOs) that won't require raw VHDL coding but using the DSP Builder development tool in Simulink to generate the code via a graphical block method which we can then use in a hardware description language (HDL) development software. We will show development of basic MOs used for both binary and greyscale imagery specifically erosion and dilation for both cases before moving on to develop the more advanced MOs using the basic ones previously developed as a reference point again for both binary and greyscale imagery and will hence look at opening, closing, openclose and closeopen filters in both cases as well as thinning and convex hull for binary. Testing will involve pixel by pixel simulations and whole image simulations to prove that the MOs developed work at both these levels. The testing will include Quartus II vector waveform pixel analysis, timing layouts, Simulink data-stream imagery and ModelSim analysis of produced code.

Analysis and implementation of an efficient edge detection algorithm

Edges in grayscale digital imagery are detected by localizing the zero crossings of filtered data. To achieve this objective, truncated time- or frequency-sampled forms (FSFs) of the Laplacian-of-Gaussian (LOG) filter are employed in the transform domain. Samples of the image are transformed using the discrete symmetric cosine transform prior to adaptive filtering and the isolation of zero crossings. This paper evaluates an adaptive blockwise filtering procedure, based on the FSF of the LOG filter, which diminishes the edge localization error, emphasizes the signal-to-noise ratio around the edge, and extends easily to higher dimensions. Theoretical expressions and applications to digital images are presented.

Distributed surveillance and reconnaissance using multiple autonomous ATVs: CyberScout

The objective of the CyberScout project is to develop an autonomous surveillance and reconnaissance system using a network of all-terrain vehicles. We focus on two facets of this system: 1) vision for surveillance and 2) autonomous navigation and dynamic path planning. In the area of vision-based surveillance, we have developed robust, efficient algorithms to detect, classify, and track moving objects of interest (person, people, or vehicle) with a static camera. Adaptation through feedback from the classifier and tracker allow the detector to use grayscale imagery, but perform as well as prior color-based detectors. We have extended the detector using scene mosaicing to detect and index moving objects when the camera is panning or tilting. The classification algorithm performs well with coarse inputs, has unparalleled rejection capabilities, and can flag novel moving objects. The tracking algorithm achieves highly accurate (96%) frame-to-frame correspondence for multiple moving objects in cluttered scenes by determining the discriminant relevance of object features. We have also developed a novel mission coordination architecture, CPAD (Checkpoint/Priority/Action Database), which performs path planning via checkpoint and dynamic priority assignment, using statistical estimates of the environment's motion structure. The motion structure is used to make both preplanning and reactive behaviors more efficient by applying global context. This approach is more computationally efficient than centralized approaches and exploits robot cooperation in dynamic environments better than decoupled approaches.

SOAR: System of associative relations

An associative memory architecture for modeling pairwise pixel interactions in imagery is proposed. More critically, a novel clustering algorithm and a similarity measure for segmentation of gray level images are developed. The inter-pixel interactions are modeled by the direction of the first-order derivative computed over several tokens exploiting the underlying inter-pixel associations. The proposed similarity measure compares two patterns based on the relative ordering of pixel values rather than the gray values themselves, and thus, is invariant to gray level biases. SOAR similarity criterion has been observed to be tightly coupled with the ubiquitous mean-square error distance measure. SOAR clustering algorithm, which encodes inter-pixel associations in an ensemble of image blocks resulted in visually consistent segmentation of ball grid array (BGA) images. SOAR, as an auto-associative memory, has been observed to be superior to the existing techniques using binary neural networks for associative storage of gray scale imagery in both the computational complexity and the processing unit requirements. Our simulations show that SOAR is capable of restoring missing parts of a learned (stored) gray level image using its regenerative capability. SOAR is parallel in nature and can be implemented using a synchronous or asynchronous parallel system with no added complexity.

Image binarization techniques for correlation-based pattern recognition

Correlation using binary images is suited to efficient digital realization or convenient optical implementation. Binarization algorithms are required in order to match grayscale imagery to these binary corre- lation architectures. We present several novel point-wise and block-wise binarization techniques all of which outperform the grayscale matched filter for large values of input signal-to-noise ratio (SNR50 dB). We dis- cuss direct binarization methods based on global thresholds, local thresholds, histogram equalization, edge-enhancement, and statistical binarization, as well as indirect methods based on auto- and cross- correlation techniques. These point-wise methods are shown to offer poor noise tolerance and a new block-wise binarization method is intro- duced to enhance recognition at low values of SNR. This block-wise technique is motivated by vector quantization-based image compression and offers performance superior to the grayscale matched filter for an input SNR as low as 212 dB. © 1999 Society of Photo-Optical Instrumentation Engineers. (S0091-3286(99)02611-2)

Transform coding of stereo image residuals

Stereo image compression is of growing interest because of new display technologies and the needs of telepresence systems. Compared to monoscopic image compression, stereo image compression has received much less attention. A variety of algorithms have appeared in the literature that make use of the cross-view redundancy in the stereo pair. Many of these use the framework of disparity-compensated residual coding, but concentrate on the disparity compensation process rather than the post compensation coding process. This paper studies specialized coding methods for the residual image produced by disparity compensation. The algorithms make use of theoretically expected and experimentally observed characteristics of the disparity-compensated stereo residual to select transforms and quantization methods. Performance is evaluated on mean squared error (MSE) and a stereo-unique metric based on image registration. Exploiting the directional characteristics in a discrete cosine transform (DCT) framework provides its best performance below 0.75 b/pixel for 8-b gray-scale imagery and below 2 b/pixel for 24-b color imagery, In the wavelet algorithm, roughly a 50% reduction in bit rate is possible by encoding only the vertical channel, where much of the stereo information is contained. The proposed algorithms do not incur substantial computational burden beyond that needed for any disparity-compensated residual algorithm.

DCT transform coding of stereo images for multimedia applications

One form of multimedia finding increasing utility in factory automation is stereo imaging. Its uses include remote operation, telepresence, and object positioning. Stereo imaging, with doubled frames, requires high data-rate collection and transmission. Image compression techniques can be utilized to reduce the transmission bandwidth and/or storage space requirements of the stereo pair. This paper develops image compression techniques specific to stereo imaging and compares the performance with nonstereo methods. Disparity compensated residual image coding exploits the redundancy between the two images in a stereo pair. Stereo residuals possess special features that can be exploited within a discrete cosine transform (DCT) image coding framework. Use of these features results in several decibels of performance enhancement across a range of scene types. This approach provides its best performance below 0.75 b/pixel bitrate for 8-b grayscale imagery and below 2 b/pixel for 24-b color imagery.

Implementation of mathematical morphological operations for spatial data processing

Mathematical morphology is a theory of image transformations which is based on set-theoretical, geometrical, and topological concepts. The methodology is useful particularly for the analysis of the geometrical structure in an image. Since many data types in the geosciences are related to spatial location, it is expected that the mathematical morphological operations are particularly important for geoscientific spatial data processing. In this paper, the binary mathematical morphological operations of dilation, erosion, opening, closing, hit/miss transform, and thinning are implemented using the C programming language and the PCIDSK C Toolbox. Dilation, erosion, opening, and closing operations are extended to gray-scale imagery. As an example, gray-scale image dilation and erosion operations are applied to an aeromagnetic image from the File Lake area, Manitoba, Canada for edge and linear feature detection. The detected edge and linear features are scaled into a binary image through thresholding, which allows for further binary morphological operations. Thinning operations are applied to binary images to obtain connected one-pixel thick skeletons. Trimming operations are tested to remove skeletal legs resulting from the thinning operation. Results from both binary and gray-scale images show that mathematical morphological operations can be applied efficiently to the processing of spatial data in the geosciences. The thinning operation makes it possible for edge and linear features to be vectorized and put directly into a geographic information system (GIS).

Vehicle detection and recognition in greyscale imagery

This paper details a novel two-stage vehicle detection and recognition algorithm by combining an image-processing region of interest (ROI) designator to cue a secondary recognition process implemented using principal component analysis (PCA) as input to a Multi-Layered Perceptron (MLP) classifier. Both the image-processing system and the MLP classifier have been designed for real-time implementation and data-fusion with other information sources.

Optimal closed boundary identification in gray-scale imagery

Identification of closed boundary contours is an important problem in image analysis because boundaries delineate the structural components, or objects, present in a scene. Most filter-based edge-detection methods do not have a mechanism to identify a group of edge sites that defines a complete closed object boundary. In this paper, we construct a suitable parameter space of one-pixel-wide closed boundaries for gray-scale images that reduces the complexity of the boundary identification problem. An algorithm based on stochastic processes and Bayesian methods is presented to identify an optimal boundary from this space. By defining a prior probability model and appropriately specifying transition probability functions on the space, a Markov chain Monte Carlo algorithm is constructed that theoretically converges to a statistically optimal closed boundary estimate. Moreover, this approach ensures that implementation via computer will result in a final boundary estimate that has the necessary property of closure which previous stochastic approaches have been unable to achieve.

Vehicle Detection and Recognition in Greyscale Imagery

This paper details a novel two-stage vehicle detection and recognition algorithm by combining an image processing region of interest (ROI) designator to cue a secondary recognition process implemented using principal component analysis (PCA) as input to a Multi-Layered Perceptron (MLP) classifier. Both the image processing system and MLP classifier have been designed for reed-time implementation and data-fusion with other information sources.

Noise immunity of threshold decomposition optoelectronic order-statistic filtering

High-speed analog optoelectronic image processing systems are generally characterized by low accuracy and are not well suited to applications requiring large dynamic range. We show that the low accuracy of a typical analog optoelectronic convolver can be tolerated in a high-accuracy application when threshold decomposition is used to perform order-statistic processing on digitized gray-scale imagery. The robustness of the method in the presence of system noise, combined with the parallelism of the optical convolver, suggests that hybrid analog optical/digital electronic processors for order-statistic image processing may be attractive alternatives to all-electronic processors. We analyze the immunity of such a hybrid processor to noise generated by spatial variations of emitter and photodetector arrays for the order-statistic operations of median filtering, dilation, and erosion.

A quantitative luminescence imaging system for biochemical diagnostics

A prototype quantitative low-light-level imaging system, capable of providing calibrated gray-scale imagery of radio-frequency (rf) stimulated chemiluminescence distributions in biological samples, has been developed by these authors. The quantitative luminescence imaging system (QLIS) represents a significant advance in the experimental instrumentation used for the study of such reactions, in that the system allows time-resolved and spatially resolved quantitative analyses to be performed. Measurements of chemiluminescence have traditionally been made by using individual photon counting photomultiplier tubes, integrating over the spatial volume and over time. The QLIS provides imagery at video rates, in which individual or integrated multiple frames can be calibrated to give a quantified image of the two-dimensional distribution of chemiluminescence in the sample. The system has a temporal resolution of 1/30th of a second, spatial resolution of 0.25 mm, a 40-mm-diam linear field of view, and a radiometric sensitivity range of 10−3–10−8 W/m2 sr. Radiometric calibration of the video imagery is accomplished with a self-luminous light source of known radiance and spectral content. The concept can be implemented for measurement at three scales: using a microscope for microscopic-scale measurements; using a wire mesh rf waveguide for medium-scale samples contained in a flat, cylindrical sample cell; or in an anechoic chamber for small- to human-body-sized samples. The prototype system has been implemented at the medium scale, with a wire mesh waveguide. Efforts are under way for development of the other two systems.