By Topic

Image Analysis and Interpretation, 1998 IEEE Southwest Symposium on

Date 5-7 April 1998

Filter Results

Displaying Results 1 - 25 of 48
  • 1998 IEEE Southwest Symposium on Image Analysis and Interpretation (Cat. No.98EX165)

    Save to Project icon | Request Permissions | PDF file iconPDF (386 KB)  
    Freely Available from IEEE
  • Author index

    Page(s): 263
    Save to Project icon | Request Permissions | PDF file iconPDF (72 KB)  
    Freely Available from IEEE
  • Lexicodes in the space of foot patterns for image classification

    Page(s): 97 - 102
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (532 KB)  

    In this paper we extend work presented in Ashlock and Davidson (1997) on the automatic classification of textures with foot patterns. We begin by verifying that a technique suggested in the earlier research permits us to distinguish between textures which the original technique could not classify. We then define a metric on the space of the foot patterns and construct lexicodes of the foot patterns that yield a new technique for distinguishing the textures. The lexicodes of the foot patterns are used to construct vectors of entropy values in R n and a clustering algorithm on those vectors is used to classify the textures. This new technique uses much of the machinery of the original technique but is unsupervised, requiring no training examples. The results of using this unsupervised technique are very similar to the results originally obtained with the supervised algorithm, including the inability to distinguish two of the six texture types in the test set. We blend the technique for distinguishing the two similar textures with the lexicode technique with partial success. We present results on binary image data but our goal is to achieve automatic classification of any gray-value texture. This has the potential to be used in automated object recognition, image retrieval from databases, and compression and data transmission applications View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Practical implementation of multirate convolution for multiresolution image processing

    Page(s): 217 - 222
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (652 KB)  

    This paper describes key issues associated with the computation of undecimated multiresolution signal decompositions. An effective solution is obtained by a general-purpose convolution programming model that automatically accounts for filter origin displacement and the sparse nature of upsampled filters. The handling of image boundaries during convolution is also discussed in detail View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Projected mean curvature smoothing for vector-valued imagery

    Page(s): 121 - 126
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (996 KB)  

    In this note, we formulate a general modified mean curvature based equation for image smoothing and enhancement. The key idea is to consider the image as a graph in some Rn, and apply a mean curvature type motion to the graph. We consider some special cases relevant to greyscale and color images View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Document segmentation using texture variance and low resolution images

    Page(s): 164 - 167
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (352 KB)  

    This paper describes a document segmentation method based on segmentation by texture using low resolution gray level images. The method is derived from the human vision perception theory. The concepts used from this theory are, global to local processing and low resolution information. If a document is viewed at a certain distance far from a person, the person sees a blurred image of the document, but is still able to detect the different blocks of the document. Detection is possible since each block has a specific texture pattern. These patterns correspond to regions of text, regions of graphics and regions of pictures. Thus the theory to prove is that a document image can be segmented into regions of text, and regions of graphics and/or pictures using the texture of low resolution images. The method presented in this paper, despite its simplicity, has shown to be effective and robust. It was designed to work with free format documents, text in background other than white, skew greater than 10 degrees. It requires less computation than the segmentation methods using texture described in other papers View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Bidimensional retinal blood vessel reconstruction by a new color edge tracking procedure

    Page(s): 232 - 237
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (524 KB)  

    The authors present a new color edge tracking procedure in order to achieve a bidimensional reconstruction of the retinal blood vessels. The major branches of the reconstructed vessels will serve as landmarks in order to locate the retinal lesions called CytoMegaloVirus (CMV) retinitis, on the color fundus images of patients with acquired immune deficiency syndrome (AIDS). The reconstruction is based on a recursive tracking of the two vessel edges extracted by means of color edge detection. After an interactive selection of the starting edge pixels, the tracking process automatically searches for the side-branches of the major vessel being tracked, providing a tree representation of the vessel vasculature. During the tracking, contour and region information (color of the vessel body) are associated in order to get more accurate extraction View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Use of diffusion techniques for edge preservation for fractal coders

    Page(s): 65 - 69
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (588 KB)  

    Based on the correlation of self-similarity, fractal coders compress digital images by relating image blocks pairs at different scales in the images. The pairing of large to small block sizes dictates the compression ratio and the quality of the reconstructed image. Inaccurate fractal mappings at large image block sizes increase losses of edge information, discontinuities at boundaries and blocking effects. Partitioning the large block into smaller blocks overcomes these problems but significantly lowers the compression ratios. In addition, partitioning does not insure the retention of significant edges. We show how diffusion techniques can be used to overcome some of these problems while preserving significant edge information at a lower bit rate cost. By expanding the basic diffusion equation to contain a scalar function based on the edges of the original image, we can use the diffusion process to smooth along the direction of significant edges and sharpen in the direction of the edges. In this manner, we can restore edge information and smooth discontinuities and blocking effects View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Local colour image segmentation using singular value decomposition

    Page(s): 148 - 153
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (612 KB)  

    A method was developed to segment images of complex scenes based on color content. The output of an interest operator provides focus toward regions within an image to be sampled for color content. Statistics for each data set sampled are used to cluster and estimate bounded regions within a transformed color space. Each region respectively represents a specific set. Mappings to transformed regions of color space are found using the singular value decomposition. The mean and variance of each color sample in the transformed color space represent characteristic features for their sampled set of points. Color segmentation is accomplished by establishing whether image pixels belong to any subset represented by the characteristic features. This work contributes a method to color-segment targets in images using local color information within an image stream View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Segmentation strategies with multiple analysis for an SMD object recognition system

    Page(s): 59 - 64
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (528 KB)  

    We present two segmentation strategies that use multiple analysis for the preprocessing stage of an object recognition system to detect the presence of surface mounted devices (SMD) on printed circuit boards. This work concentrates only on the preprocessing stage and simple segmentation algorithms for fast real-time implementation. The system uses two images of the same scene, with top and side illuminations. One approach uses both the top- and the side-illuminated images while the other one uses only the top-illuminated image. Experiments are performed using the two model-based segmentation approaches, which produce a gray level region of interest (ROI) that has the SMD isolated as a target when it is present or a smaller area when the SMD is absent. The suppression of the copper-pads is a key step in the processing. For the first strategy, the comparison criteria used to evaluate its performance are the binary area and the energy of the ROI. For the second strategy, our evaluation is a comparison of the results with a fixed size reference ROI mask located at a centroid chosen by visual analysis of the image. Using a database of 1500 images, the distributions of the two comparison criteria are shown for three possible scenes: the SMD is present in the images, the SMD is absent but a speck of glue is present, or the SMD and the glue are both absent. Examples of the processing results are shown View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive reconstruction using multiple views

    Page(s): 47 - 52
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (816 KB)  

    This paper introduces a novel algorithm for extracting the optical flow obtained from a translating camera in a static scene. Occlusion between objects is incorporated as a natural component in a scene reconstruction strategy by first evaluating and reconstructing the foreground and then excluding its influence on the partly occluded objects behind View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • High-scale edge study for segmentation and contour closing in textured or noisy images

    Page(s): 109 - 114
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (740 KB)  

    When they are used in a high-scale way, the edge detectors based on Canny's approach provide edges which are well localized in non-noisy and non-textured areas of the image, but which are too numerous in other areas. The aim of this paper is to study these high-scale edges, especially their local density. High-scale edges are studied by the way of features which characterize various situations such as texture areas, contours corresponding to object frontiers with or without noise. Such features may even characterize various textures. These features may be used to build a map of textured/noisy areas and then to perform a segmentation method cooperation. Another application is the contour closing, using distances between textures View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Compression degradation metrics for analysis of consistency in microcalcification detection

    Page(s): 35 - 40
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (708 KB)  

    The authors motivate, define and illustrate the behavior of four scalar metrics which measure the degree to which microcalcification (μuCa++) detection is affected by lossy compression distortion. The JPEG method is used at various compression rates determined by a single parameter. The authors' metrics are specifically targeted to measure the degree of invariance of the processing to the presence of a compression-decompression step and not absolute processing success. The detection of the μCa++s is done with a two-step process of selective median filtering and wavelet domain hard thresholding with saturation. This latter step produces a binary image revealing small structures which are the desired objects in the image. The authors have also chosen to perform the enhancement before the compression step as suggested by their previous work to allow higher compression rates. In the system used to do the comparison, the two branches of processing are identical except for the compression/decompression step and their results are binary images. The inputs to the metric computation system are these two images which produce four scalar metrics which are various percentages of differences in pixel and region activity. Using seven image sections from mammograms, the authors illustrate the results at various compression ratios in the range 28:1 to 42:1 for each image and also averaged over the set of seven. The three pixel based metrics decrease monotonically as the JPEG compression quality parameter increases to reduce the compression ratio. The region based metric does not change monotonically due to the more complex nature of region analysis View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Color image segmentation using multi-scale clustering

    Page(s): 142 - 147
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (540 KB)  

    The use of clustering in color image segmentation poses two distinct problems: (a) equal distances throughout a color space may not be perceived equally by the human visual system, and (b) the number of color clusters must be predetermined. This paper describes a color clustering method that resolves these problems. The first problem is addressed by operating in the nonlinear, geodesic chromaticity space where color shifts are nearly uniform. The second problem is remedied by utilizing a newly developed multi-scale clustering algorithm. This algorithm determines the prominent numbers of color clusters via an objective measure named lifetime. The obtained segmentation results indicate that this color segmentation approach identifies the prominent color structures or objects in a color image View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Texture classification using combined feature sets

    Page(s): 103 - 108
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (536 KB)  

    We consider two methods to combine texture descriptions for classification: a composite feature vector which combines data additively, and an extended k-nearest-neighbour (KNN) rule which returns a decision based on the highest confidence in features, both aimed to improve classification capability. These have been used to combine a wide range of relatively simple texture features, and have been shown to have significant advantage. Although nearly all previous approaches have used a limited subset of the Brodatz database, the new techniques have been applied to the whole Brodatz database with evaluation independent of the number of test classes used by measuring the number of perfect classes. The results of these new methods of combination show that an overall classification rate exceeding 90% can be achieved with 71 perfect classes, improving capabilities above using the measures individually View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Morphological and related image processing techniques for the detection of microcalcifications in mammographic images

    Page(s): 29 - 34
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (608 KB)  

    This paper describes a simple technique for locating microcalcifications in digital mammograms. These methods use the fact that many smoothing techniques are never applied to mammograms as they erase the important features. This is turned to an advantage by subtracting the smoothed image from the original, thus leaving only the important features. A variety of smoothers and enhancers are applied to two image data sets, and the results reported View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Handwritten digit recognition using combination of neural network classifiers

    Page(s): 168 - 173
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (484 KB)  

    A new classification scheme for handwritten digit recognition is proposed. The method is based on combining the decisions of two multilayer perceptron (MLP) artificial neural network classifiers operating on two different feature types. The first feature set is defined on the pseudo Zernike moments of the image whereas the second feature type is derived from the shadow code of the image using a newly defined projection mask. A MLP network is employed to perform the combination task. The performance is tested on a data base of 15000 samples and the advantage of the combination approach is demonstrated View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Texture defect detection using subband domain co-occurrence matrices

    Page(s): 205 - 210
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (708 KB)  

    In this paper, a new defect detection algorithm for textured images is presented. The algorithm is based on the subband decomposition of gray level images through wavelet filters and extraction of the co-occurrence features from the subband images. Detection of defects within the inspected texture is performed by partitioning the textured image into non-overlapping subwindows and classifying each subwindow as defective or nondefective with a mahalanobis distance classifier being trained on defect free samples a priori. The experimental results demonstrating the use of this algorithm for the visual inspection of textile products obtained from the real factory environment are also presented View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Removal of image defocus and motion blur effects with a nonlinear interpolative vector quantizer

    Page(s): 1 - 5
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1424 KB)  

    In this paper, results are presented which demonstrate the removal of image defocus and motion blur effects using an algorithm based on nonlinear interpolative vector quantization (NLIVQ). The algorithm is trained on original and diffraction-limited image pairs which are representative of the class of images of interest. The discrete cosine transform is used in the code-book design process to control complexity. Imagery processed with this algorithm demonstrate both qualitative and quantitative improvements (as measured by the peak signal-to-noise-ratio before and after processing) View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A robust registration technique for multi-sensor images

    Page(s): 87 - 90
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (552 KB)  

    This paper describes an improved technique to register multi-sensor images by segmenting the images by adaptive clustering prior to performing preprocessing and cepstrum operation to determine the translational displacement. The difficulty in registering multi sensor images lies in the fact that the images of the same scene acquired by different sensors often appear different in detailed structures. Therefore the common features existing in such images need to be identified by suitable preprocessing operations for the success of the cepstral registration technique. Experimental results demonstrate the feasibility of successful cepstral registration of SAR and electro-optic images of the same scene despite apparent noticeable differences in some embedded structures thus providing a potential powerful tool for automated registration View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Design of a computer vision based tree ring dating system

    Page(s): 256 - 261
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (688 KB)  

    The purpose of this paper is to describe the design and implementation of a computer vision based analysis system for dendrochronology. The issues involved in the detection and analysis of tree rings are not unique to the application, but are likely of interest to anyone developing automated image analysis systems View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Scene change detection in MPEG domain

    Page(s): 12 - 17
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (528 KB)  

    Video is an important and challenging media and requires sophisticated indexing schemes for efficient retrieval from visual databases. Video segmentation is a fundamental step in video indexing and involves detection of scene changes. In this paper, we propose a fast and robust algorithm for detecting video shot boundaries in the MPEG-2 compressed bitstream with minimal decoding View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Template construction and matching for identification of cells in differential interference contrast microscope images

    Page(s): 238 - 243
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (944 KB)  

    Template matching, for identification of cells in differential interference contrast (DIC) microscope images by digital image analysis, is complex. This is due to the nature of DIC images, namely an apparent illumination effect across cells in the image. Different cell sizes and orientations cause further complications. Three methods of template construction are proposed for different types of cell, and a successful automatic method for cell counting and sizing is described. These methods compensate for the complexities in the images View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Image reconstruction using the Viterbi algorithm

    Page(s): 160 - 163
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (424 KB)  

    Many systems in widespread use concentrate on the imaging of binary objects, e.g., the archival storage of text documents on microfilm or the facsimile transmission of text. Due to the imperfect nature of such systems, the binary image is unavoidably corrupted by blur and noise to form a grey-scale image. We present a technique to reverse this degradation which maps the binary object reconstruction problem into a Viterbi state-trellis. We assign states of the trellis to possible outcomes of the reconstruction estimate and search the trellis in the usual optimal fashion. Our method yields superior estimates of the original binary object over a wide range of signal-to-noise ratios (SNR) when compared with conventional Wiener filter (WF) estimates. For moderate blur and SNR levels, the estimates produced approach the maximum likelihood (ML) bound on estimation performance View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Modified mean curvature motion for multispectral anisotropic diffusion

    Page(s): 154 - 159
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1376 KB)  

    This paper introduces a new anisotropic diffusion algorithm for enhancing and segmenting multispectral image data. The algorithm is based upon mean curvature motion. Using a modified image gradient computation, the diffusion method is further improved by allowing the control of feature scale, and the sensitivity to heavy-tailed noise is eliminated. For comparison, a vector distance dissimilarity method is introduced and extended for multi-scale processing. The experiments on remotely sensed imagery and color imagery demonstrate the performance of the algorithms in terms of image entropy reduction and impulse elimination as well as visual quality View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.