Turbo segmentation of textured images

We consider the problem of semi-supervised segmentation of textured images. Existing model-based approaches model the intensity field of textured images as a Gauss-Markov random field to take into account the local spatial dependencies between the pixels. Classical Bayesian segmentation consists of also modeling the label field as a Markov random field to ensure that neighboring pixels correspond to the same texture class with high probability. Well-known relaxation techniques are available which find the optimal label field with respect to the maximum a posteriori or the maximum posterior mode criterion. But, these techniques are usually computationally intensive because they require a large number of iterations to converge. In this paper, we propose a new Bayesian framework by modeling two-dimensional textured images as the concatenation of two one-dimensional hidden Markov autoregressive models for the lines and the columns, respectively. A segmentation algorithm, which is similar to turbo decoding in the context of error-correcting codes, is obtained based on a factor graph approach. The proposed method estimates the unknown parameters using the Expectation-Maximization algorithm.     

Adaptive segmentation of noisy and textured images

An image segmentation algorithm is described which is based on the integration of signal model parameter estimates and maximum a posteriori labelling. The parameter estimation is based on either a maximum likelihood-based method for a quadric signal model or a maximum pseudo-likelihood based method for a Gauss-Markov signal model. The first case is applicable to standard grey-level image segmentation as well as segmentation of shaded 3D surfaces, while the second case is applicable to texture segmentation. A key aspect of the algorithm is the incorporation of a coarse to fine processing strategy which limits the search for the optimum labelling at any one resolution to a subset of labellings which are consistent with the optimum labelling at the previous coarser resolution. Consistency is in terms of a prior label model which specifies the conditional probability of a given label in terms of the labelling at the previous level of resolution. It is shown how such an approach leads to a simple relaxation procedure based on local pyramid node computations. An extension of the algorithm is also described which performs accurate inter-region boundary placement using a step-wise refinement procedure based on a simple adaptive filter. The problem of automatic determination of the number of regions is also addressed. It is shown how a simple agglomerative clustering idea, again based on pyramid node computations, can effectively solve this problem.     

Multiple resolution segmentation of textured images

A multiple resolution algorithm is presented for segmenting images into regions with differing statistical behavior. In addition, an algorithm is developed for determining the number of statistically distinct regions in an image and estimating the parameters of those regions. Both algorithms use a causal Gaussian autoregressive model to describe the mean, variance, and spatial correlation of the image textures. Together, the algorithms can be used to perform unsupervised texture segmentation. The multiple resolution segmentation algorithm first segments images at coarse resolution and then progresses to finer resolutions until individual pixels are classified. This method results in accurate segmentations and requires significantly less computation than some previously known methods. The field containing the classification of each pixel in the image is modeled as a Markov random field. Segmentation at each resolution is then performed by maximizing the a posteriori probability of this field subject to the resolution constraint. At each resolution, the a posteriori probability is maximized by a deterministic greedy algorithm which iteratively chooses the classification of individual pixels or pixel blocks. The unsupervised parameter estimation algorithm determines both the number of textures and their parameters by minimizing a global criterion based on the AIC information criterion. Clusters corresponding to the individual textures are formed by alternately estimating the cluster parameters and repartitioning the data into those clusters. Concurrently, the number of distinct textures is estimated by combining clusters until a minimum of the criterion is reached     

Segmentation of document images

Several methods for segmentation of document images (maps, drawings, etc.) are explored. The segmentation operation is posed as a statistical classification task with two pattern classes: print and background. A number of classification strategies are available. All require some prior information about the distribution of gray levels for the two classes. Training (either supervised or unsupervised) is employed to form these initial density estimates. Automatic updating of the class-conditional densities is performed within subregions in the image to adapt these global density estimates to the local image area. After local class-conditional densities have been obtained, each pixel is classified within the window using several techniques: a noncontextual Bayes classifier, Besag's classifier, relaxation, Owen and Switzer's classifier, and Haslett's classifier. Four test images were processed. In two of these, the relaxation method performed best, and in the other two, the noncontextual method performed best. Automatic updating improved the results for both classifiers     

Segmentation of SAR images

This paper presents a new algorithm for segmentation of SAR images based on threshold estimation using the histogram. The speckle distribution in the SAR image is modeled by a Gamma function. Thus, the SAR image histogram exhibits a combination of Gamma distributions. The maximum likelihood technique is therefore used to estimate the histogram parameters. This technique requires knowledge of the number of modes of the histogram, the number of looks of the SAR image, and the initial parameters of the histogram. The second derivative of the histogram is used to estimate the number of modes. We use two methods to estimate the number of looks. Initial parameters are estimated at the maximum of the Gamma function. Thresholds are selected at the valleys of a multi-modal histogram by minimizing the discrimination error between the classes of pixels in the image. The algorithm is applied to several RADARSAT SAR images with different number of looks. The results obtained are promising.     

Segmentation of chromatic images

When analyzing color pictures one often requires that the image be segmented into meaningful regions based upon the color characteristics of the scene. Such a problem can assume two different forms. The first variation arises when particular color space characteristics are known and the goal is to detect and extract image regions which possess the given color characteristics. The second case arises when there is no a priori knowledge about the color space characteristics of the scene and the goal is to segment the scene into meaningful regions which possess uniform color space characteristics. This paper describes an interactive system which uses a decision surface modeling approach to solve the first case and uses clustering techniques in the three-dimensional color space to solve the second case. A set of examples is presented and the performance of the system is evaluated.     

Segmentation and classification of range images

The recognition of objects in three-dimensional space is a desirable capability of a computer vision system. Range images, which directly measure 3-D surface coordinates of a scene, are well suited for this task. In this paper we report a procedure to detect connected planar, convex, and concave surfaces of 3-D objects. This is accomplished in three stages. The first stage segments the range image into ``surface patches' by a square error criterion clustering algorithm using surface points and associated surface normals. The second stage classifies these patches as planar, convex, or concave based on a non-parametric statistical test for trend, curvature values, and eigenvalue analysis. In the final stage, boundaries between adjacent surface patches are classified as crease or noncrease edges, and this information is used to merge compatible patches to produce reasonable faces of the object(s). This procedure has been successfully applied to a large number of real and synthetic images, four of which we present in this paper.     

Segmentation and tracking of piglets in images

An algorithm was developed for the segmentation and tracking of piglets and tested on a 200-image sequence of 10 piglets moving on a straw background. The image-capture rate was 1 image/140 ms. The segmentation method was a combination of image differencing with respect to a median background and a Laplacian operator. The features tracked were blob edges in the segmented image. During tracking, the piglets were modelled as ellipses initialised on the blobs. Each piglet was tracked by searching for blob edges in an elliptical window about the piglet's position, which was predicted from its previous two positions.     

Segmentation of images of license plates

An approach to the segmentation of license plates is considered. A specific feature of this approach is the use of a sequence of images. An algorithm for the segmentation and recognition of license plates based on dynamic programming is proposed.     

Segmentation of images having unimodal distributions

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.     

Modeling and segmentation of noisy and textured images using gibbs random fields

This paper presents a new approach to the use of Gibbs distributions (GD) for modeling and segmentation of noisy and textured images. Specifically, the paper presents random field models for noisy and textured image data based upon a hierarchy of GD. It then presents dynamic programming based segmentation algorithms for noisy and textured images, considering a statistical maximum a posteriori (MAP) criterion. Due to computational concerns, however, sub-optimal versions of the algorithms are devised through simplifying approximations in the model. Since model parameters are needed for the segmentation algorithms, a new parameter estimation technique is developed for estimating the parameters in a GD. Finally, a number of examples are presented which show the usefulness of the Gibbsian model and the effectiveness of the segmentation algorithms and the parameter estimation procedures.     

Unsupervised segmentation of textured images by edge detection inmultidimensional feature

An algorithm for unsupervised texture segmentation is developed that is based on detecting changes in textural characteristics of small local regions. Six features derived from two, two-dimensional, noncausal random field models are used to represent texture. These features contain information about gray-level-value variations in the eight principal directions. An algorithm for automatic selection of the size of the observation windows over which textural activity and change are measured has been developed. Effects of changes in individual features are considered simultaneously by constructing a one-dimensional measure of textural change from them. Edges in this measure correspond to the sought-after textural edges. Experiments results with images containing regions of natural texture show that the algorithm performs very well     

A genetic algorithm for MRF-based segmentation of multi-spectral textured images

A segmentation approach based on a Markov random field (MRF) model is an iterative algorithm; it needs many iteration steps to approximate a near optimal solution or gets a non-suitable solution with a few iteration steps. In this paper, we use a genetic algorithm (GA) to improve an unsupervised MRF-based segmentation approach for multi-spectral textured images. The proposed hybrid approach has the advantage that combines the fast convergence of the MRF-based iterative algorithm and the powerful global exploration of the GA. In experiments, synthesized color textured images and multi-spectral remote-sensing images were processed by the proposed approach to evaluate the segmentation performance. The experimental results reveal that the proposed approach really improves the MRF-based segmentation for the multi-spectral textured images.     

On the use of a joint spatial-frequency representation for the fusion of multi-focus images

This paper shows practical examples of the application of a new image fusion paradigm for achieving a 2-D all in-focus image starting from a set of multi-focus images of a 3-D real object. The goal consists in providing an enhanced 2-D image showing the object entirely in focus. The fusion procedure shown here is based on the use of a focusing pixel-level measure. Such measure is defined in the space–frequency domain through a 1-D pseudo-Wigner distribution. The method is illustrated with different sets of images. Evaluation measures applied to artificially blurred cut and pasted regions have shown that the present scheme can provide equally or even better performance than other alternative image fusion algorithms.     

Retrieval of textured images through the use of quantization and modal analysis

This paper proposes a new method for textured image retrieval, by the modal analysis of quantized spectral point patterns as the modal correspondence method of Shapiro and Brady, to match point sets by comparing the eigenvectors of a pairwise point proximity matrix taken from the power spectrum peaks. A variant of the Carcassoni, Ribeiro and Hancock method for performing recognition is taken into account. For choosing image features to represent an image, a quantization scheme is applied. This quantization scheme acts in the spectral space given by the Fourier transform of each image. Its goal is to find a small set which represents an image efficiently, where the most important features are presented. The proposed technique is invariant to rotation and is robust in the presence of noise and damaged images. The techniques here presented are compared, and the commonly used retrieval performance measurement—precision and recall—is used as evaluation of the query results.     

Segmentation of microscope images of living cells

This paper describes a segmentation method combining a texture based technique with a contour based method. The technique is designed to enable the study of cell behaviour over time by segmenting brightfield microscope image sequences. The technique was tested on artificial images, based on images of living cells and on real sequences acquired from microscope observations of neutrophils and lymphocytes as well as on a sequence of MRI images. The results of the segmentation are compared with the results of the watershed and snake segmentation methods. The results show that the method is both effective and practical.

Segmentation of small objects in color images

A method for effective segmentation of small objects in color images is presented. It can be used jointly with region growing algorithms. Segmentation of small objects in color images is a difficult problem because their boundaries are close to each other. The proposed algorithm accurately determines the location of the boundary points of closely located small objects and finds the skeletons (seed regions) of those objects. The method makes use of conditions obtained by analyzing the change of color characteristics of the edge pixels along the direction that is orthogonal to the boundaries of adjacent objects. These conditions are generalized for the case of the well-known class of color images having misregistration artifacts. If high-quality seed regions are available, the final segmentation can be performed using one of the region growing methods. The segmentation algorithm based on the proposed method was tested using a large number of color images, and it proved to be very efficient.     

Segmentation and classification of hyperspectral images using watershed transformation

Hyperspectral imaging, which records a detailed spectrum of light for each pixel, provides an invaluable source of information regarding the physical nature of the different materials, leading to the potential of a more accurate classification. However, high dimensionality of hyperspectral data, usually coupled with limited reference data available, limits the performances of supervised classification techniques. The commonly used pixel-wise classification lacks information about spatial structures of the image. In order to increase classification performances, integration of spatial information into the classification process is needed. In this paper, we propose to extend the watershed segmentation algorithm for hyperspectral images, in order to define information about spatial structures. In particular, several approaches to compute a one-band gradient function from hyperspectral images are proposed and investigated. The accuracy of the watershed algorithms is demonstrated by the further incorporation of the segmentation maps into a classifier. A new spectral-spatial classification scheme for hyperspectral images is proposed, based on the pixel-wise Support Vector Machines classification, followed by majority voting within the watershed regions. Experimental segmentation and classification results are presented on two hyperspectral images. It is shown in experiments that when the number of spectral bands increases, the feature extraction and the use of multidimensional gradients appear to be preferable to the use of vectorial gradients. The integration of the spatial information from the watershed segmentation in the hyperspectral image classifier improves the classification accuracies and provides classification maps with more homogeneous regions, compared to pixel-wise classification and previously proposed spectral-spatial classification techniques. The developed method is especially suitable for classifying images with large spatial structures.     

Optimal linear representations of images for object recognition

Although linear representations are frequently used in image analysis, their performances are seldom optimal in specific applications. This paper proposes a stochastic gradient algorithm for finding optimal linear representations of images for use in appearance-based object recognition. Using the nearest neighbor classifier, a recognition performance function is specified and linear representations that maximize this performance are sought. For solving this optimization problem on a Grassmann manifold, a stochastic gradient algorithm utilizing intrinsic flows is introduced. Several experimental results are presented to demonstrate this algorithm.