Abstract: Publisher Summary In image compression or enhancement, the desired output is a picture— an approximation to, or an improved version of, the input picture. Another major branch of picture-processing program deals with image analysis or scene analysis. This chapter discusses picture and scene segmentation techniques. Some segmentation operations can be applied directly to any picture; others can only be applied to a picture that has already been partially segmented as they depend on the geometry of the parts that have already been extracted from the picture. For example, a chromosome picture can be (crudely) segmented by thresholding its gray level. Once this has been done, further segmentation into individual chromosomes can be attempted, based on connectedness, size, and shape criteria. There is no single standard approach to segmentation. The perceptual processes involved in segmentation of a scene by the human visual system, such as the Gestalt laws of organization, are not yet well understood.

Abstract: A gradient relaxation method based on maximizing a criterion function is studied and compared to the nonlinear probabilistic relaxation method for the purpose of segmentation of images having unimodal distributions. Although both methods provide comparable segmentation results, the gradient method has the additional advantage of providing control over the relaxation process by choosing three parameters which can be tuned to obtain the desired segmentation results at a faster rate. Examples are given on two different types of scenes.

Abstract: Recursive segmentation of an image into regions using histograms is one of the most widely used techniques for image segmentation. At CMU, several versions of a region segmentation program have been developed based on this technique (Ohlander, Price, Shafer and Kanade). Based on these experiences, this paper discusses issues of recursive region segmentation in the context of PHOENIX, the newest version of region segmentation program, running on a VAX 11/780 under UNIX. The issues discussed in this paper include: Image features to be used in histogramming; comparison of the algorithm with other techniques; important improvements made in PHOENIX over its predecessor (Ohlander and Price); and some inherent problems in histogram-based segmentation together with suggestions for minimizing them. PHOENIX is being incorporated into the ARPA Image Understanding Testbed, under construction at SRI International.

Abstract: An image segmentation algorithm based on histogram clustering and probabilistic relaxation labeling is explored. The algorithm is evaluated by means of a set of artificially generated test images with known parameters. Two sources of pixel labeling errors are revealed. The first derives from distribution overlap in the histogram and leads to fragmented or missing regions in a segmentation. The second derives from the gloal nature of the compatibility coefficients used in the relaxation process. The coefficients are shown to be insufficient to correct certain labeling errors and can even cause the destruction of fine image details during the course of the relaxation updating process. A potential solution to these problems is shown to be obtainable by using orientation dependent compatibility coefficients and localizing the scope of the algorithm to small subimages followed by a merging of the segmented subimages.

Abstract: This paper, the second of two, is a preliminary attempt to translate a theoretical framework based on the concepts of segmentation and networks of power, as outlined in the first paper, to the establishment level. The data used in the empirical analysis, which are drawn from the West Midlands ironfoundry industry, enable surveyed ironfoundries to be assigned to their appropriate segments. The linkage patterns typical of each of these segments are described and this analysis offers a reinterpretation of previous linkage studies.

Abstract: This paper presents a new segmentation and coding algorithm for nonbinary images. The algorithm performs contour coding of regions of equally valued and connected pixels. It consists of two distinct phases: raster scanning and border following. In this sense it is similar to algorithms presented by Kruse. However, the algorithm of this paper is considerably improved since it correctly segments truly nonbinary images. The basic idea of the algorithm is to "coat" (color, label) the borders (the cracks) between the regions from both sides in two separate border-following procedures called island following and object following. Thus, all adjacencies between the objects are systematically explored and noted. Furthermore, the raster scanner, which exhaustively searches the image for new regions, can easily determine from existing/nonexisting coating which boundaries have been traced out and which have not. The algorithm can be considerably simplified for the binary image case.

Abstract: : PHOENIX is a computer program for segmenting images into homogeneous closed regions. It uses histogram analysis, thresholding and connected-components analysis to produce a partial segmentation, then resegments each region until various stopping criteria are satisfied. Its major contributions over other recursive segmenters are a sophisticated control interface, optional use of more than one histogram-dependent intensity threshold during tentative segmentation of each region. and spatial analysis of resulting subregions as a form of "look-ahead" for choosing between promising spectral features at each step. PHOENIX was contributed to the DARPA Image Understanding Testbed at SRI by Carnegie-Mellon University. This report summarizes application for which PHOENIX is suited, the history and nature of the algorithm, details of the Testbed implementation, the manner in which PHOENIX is invoked and controlled, the type of results that can be expected, and suggestions for further development. Baseline parameter sets are given for producing reasonable segmentations of typical imagery.

Abstract: Market segmentation, when used as a method for accounting for cross-sectional taste differences, is often applied in travel-demand analyses. This paper suggests the employment of the life-style concept as an improved basis for segmentation. Life-style is defined as the behavioral pattern that results from three major life decisions: the decision to form a household, the decision to participate in the labor force, and the orientation toward leisure. By using available socioeconomic variables, an attempt is made to identify life-style groups and to use them as market segments in a joint mode and destination choice model. Two tests are presented. One is the use of life-style-specific variables in the model specification and the other is the estimation of separate models for each market segment. Both approaches have shown an improvement in the model performance compared with either a pooled model or an income-based and a life-cycle/occupation-based segmentation. Further refinement of the ability to identify life-styles is suggested.

Abstract: : Several segmentation techniques were applied to a set of 51 FLIR (forward-Looking Infrared) images of four different types, and the results were compared to hand segmentations. There were substantial differences in performance, indicating that the choice of proper technique is very important. The segmentation techniques used were 'superslice,' 'pyramid spot detection', two versions of 'relaxation', pyramid linking', and 'superspike', One technique, 'superspike', outperformed all the others, detecting 88% of the targets and yielding only 1.6 false alarms per true target.

Abstract: A now method of piece-wise linear functional approximation of planar curves is introduced. This method is baaed on an adaptive segmentation procedure that alleviates the need for segment number estimation and error norm minimization. It is shown that this approach results in a higher computational efficiency than existing methods with comparable accuracy

Abstract: Aiming at designing the image processing unit of a visual prosthesis for sight handicapped, an efficient picture simplification scheme for tactile outputs is proposed. Some psychological considerations are given to help in its development. A new simple and robust segmentation method which combines the advantages of both the classical edge and region approaches is introduced. This method uses, on the one hand, the inherent property of edges to provide transition thresholds between regions of monotonous intensity, and on the other hand, the fact that region segmentation methods give closed regions. A region labeling is applied using those thresholds, yielding well outlined areas. Artificial textures are introduced to help in the tactile discrimination of shapes.

Abstract: Dynamic programming has been shown to give excellent results for isolated word recognition Two major drawbacks to dynamic time warping are the excessive storage and computational requirements for large vocabularies This paper proposes a technique of data reduction using acoustic segmentation which gives savings in both these areas To demonstrate the advantages of this technique three distance measures are evaluated for performance in the unsegmented case The one yielding the best recognition accuracy is used to compare three algorithms for matching segmented templates Performance in terms of recognition accuracy, computation time, and savings in storage are given

Abstract: Columnar architecture is a well established organizational principle for a variety of cortical systems. If two topographically mapped receptor systems, which receive slightly different "views" of the same physical stimulus, are interlaced as "columns", then the difference map of the afferent inputs is coded within a spatial frequency channel of the resultant map. The difference map of the left and right retinal views of a three dimensensional scene contains cues for the binocular disparity of the objects in the scene. Physical objects which are located at a common distance from the observer will be represented by area's of difference mapping which possesss common cortical textural values. Thus, segmentation of the cortical representation of the visual scene by values of positional disparity may be accomplished by conventional monocular segmentation techniques, applied to the cortical representation. The difference map is carried by a spatial frequency modulation determined by the period of the columnar interlacing. Ocular dominance columns in human striate cortex suggest a spatial frequency carrier which is roughly equal to the inverse of Panum's area. Since the difference mapping is a global attribute of the cortical representation, and is not contingent on the existence of labeled single cell feature extractors, the difference mapping algorithm represents a distinct alternative to conventional single cell approaches to feature extraction. The difference mapping algorithm is briefly discussed in relation to other difference channels, such as color opponent segmentation and binocular orientation disparity. It is suggested that difference mapping may reflect a general synergistic mechanism relating topographic mapping and columnar architecture, which reduces the problem of feature extraction and segmentation for depth and color opponent channels to a single "textural" mechanism.

Abstract: This correspondence presents a procedure to recognize handprinted alphanumeric characters written on a graphic tablet. After preprocessing, the input character is segmented into a polygon using a simple segmentation procedure. A feature vector is formed by the parameters which describe the segments of the polygon. Classification is done in two steps, the first one based on structural information extracted from the feature vector and the second based on statistical decision rule using parameters of the segments. A recursive learning procedure is introduced in the statistical classifier. The evaluation includes the measurement of recognition rates using several statistical classifiers, the validity test on the hypothesis concerning the distribution of feature vectors and the possibility of further simplification using principal axis analysis. Databases were created and used for the evaluation.

Abstract: A correct segmentation of cell images intonucleus, cytoplasm and background is a prerequi- site for a working automatic pre- screening devicefor cervical cytology. This paper presents an al- gorithm for determining the segmentation thres-holds. It is based on a- priori assumptions about a cell's shape and size and works on one object at a time, disregarding everything else in the image.The algorithm is capable of verifying that theisolated object really looks like a cell; an es-sential feature in an automatic system. The nu-cleus and cytoplasm thresholds are decided upon almost independently of each other. The algorithmworks by tracking iso- density contours around theobject to be isolated and its execution time is thus proportional to the length of the contourrather than the area of the image. Some prelimi-nary results are given and the possibility of ef-ficiently implementing the algorithm in hardwareis discussed.INTRODUCTION A lot of effort has been put into the re-search towards an automated cervical pre- screeningdevice based on image processing. One reason whythese attempts so far

Abstract: A method for segmenting aerial images using edge information to create regions of similar or smoothly varying intensity is discussed. Region segmentation using edges directly as input cannot be successful because boundaries are seldom perfectly closed. In the present method, we preprocess the edge image to close gaps and create a binary image from which we extract the connected regions. We compare the results obtained with this method and a traditional region-splitting method for 2 different views of an aerial scene.

Abstract: In this article the problem of pitch period detection is discussed. The following statements concerning the nature of the problem are made: (1) Pitch period detection is a segmentation problem (in the technical sense of segment). (2) 'Significant points' are required as boundary points for the definition of segments. (3) There exists a visual recognition mode for segments in the speech signal. This implies the recognition of 'significant points'. (4) The visually recognized segments may be related to auditorily recognized segments. Then an algorithm ('skeletization') is presented, which selects 'significant' points. The 'significance' of these points is examined. Characteristic properties of the suggested solution are discussed.

Abstract: A study was conducted to investigate the generality of behavioral segmentation phenomena that have been observed in single-actor situations to dyadic situations. Changes in subjects' segmentation strategies via instructional manipulations and across time were found. The generality of the feature change hypothesis from single-actor situations to dyadic-actor situations was addressed.

Abstract: : The problem of partitioning a time-series into segments is considered. The segments fall into classes, which may correspond to phases of a cycle (recession, recovery, expansion in the business cycle) or to portions of a signal obtained by scanning (background/ clutter, target, background/clutter again, another target, etc.), or normal tissue, tumor, normal tissue in medical applications. A probability distribution is associated with each class of segment. Parametric families of distributions are considered, a set of parameter values being associated with each class. With each observation is associated an unobservable label, indicating from which class the observation arose. The label process is modeled as a Markov chain. Segmentation algorithms are obtained by applying a method of iterated maximum likelihood to the resulting likelihood function. In this paper special attention is given to the situation in which the observations are conditionally independent, given the labels. A numerical example is given. Choice of the number of classes, using Akaike's information criterion (AIC) for model identification, is illustrated. Similar ideas are applied to the problem of segmenting digital images, where possible applications include SEASAT (and LANDSAT) multi-spectral images. (Author)

Abstract: Two contributions to recent issues of Papers in Japanese Linguistics have raised the question of the identity of the -rthat appears in the conjugation of Modern Japanese verbs that are based on English loan words, such as demoru 'to demonstrate (politically)' based on demo 'demonstration' , a shortened form of demonsutoreesyon. Ashworth and Lincoln (1973) claim that the -ris part of inflecs tional endings, proposing the following segmentation for forms of demoru's paradigm: (1) (Non-past) Indicative demo-ru. (2) (Non-past) Presumptive demo-roo (3) Infinitive demo-ri (4) Provisional demo-reba (5) Negative demo-ra(-}nai (6) Past (Indicative) demo-tta (Ashworth and Lincoln actually postulate /demo-ri-ta/) etc.

Abstract: : It is shown how the ordinal image segmentation problem fits into an earlier well-developed model for hierarchical clustering. Certain techniques suggested by this model are investigated and implemented on real data. The results are compared to those achieved by segmentation techniques that involve region mergers based on various notions of scatter. The techniques are examined from both an order theoretic and statistical viewpoint.