Gray Level Information Research Articles

An Interactive Technique for Cartographic Feature Extraction from Aerial and Satellite Image Sensors.

In this paper, an interactive technique for extracting cartographic features from aerial and spatial images is presented. The method is essentially an interactive method of image region segmentation based on pixel grey level and texture information. The underlying segmentation method is seeded region growing. The criterion for growing regions is based on both texture and grey level, where texture is quantified using co-occurrence matrices. The Kullback distance is utilised with co-occurrence matrices in order to describe the image texture, then the Theory of Evidence is applied to merge the information coming from texture and grey level image from the RGB bands. Several results from aerial and spatial images that support the technique are presented

An adaptive image Euclidean distance

The image Euclidean distance (IMED) considers the spatial relationship between the pixels of different images and can easily be embedded in existing image recognition algorithms that are based on Euclidean distance. IMED uses the prior knowledge that pixels located near one another have little variance in gray scale values, and defines a metric matrix according to the spatial distance between pixels. In this paper, we propose an adaptive image Euclidean distance (AIMED), which considers not only the prior spatial knowledge, but also the prior gray level knowledge from images. The most important advantage of the proposed AIMED over IMED is that AIMED makes the metric matrix adaptive to the content of the concerned images. Two ways of using gray level information are proposed. One is based on gray level distances, and the other is based on cosine dissimilarity of gray levels. Experiments on two facial databases and a handwritten digital database show that AIMED achieves the highest classification accuracy when it is embedded in nearest neighbor classifiers, principal component analysis, and support vector machines.

Systematic Study of the Urban Postconflict Change Classification Performance Using Spectral and Structural Features in a Support Vector Machine

The state of built-up features after their destruction, as well as the process of their rehabilitation, are assessed through the analysis of conflict and postconflict very high spatial resolution Ikonos images using a pixel-level support vector machine (SVM) learning classification approach. Different input vectors of the supervised SVM classifier are tested in order to assess the discrimination power of structural and spectral image descriptors: the use of spectral information only with (a) the panchromatic images at time t0 and t1, (b) the pan-sharpened images with the multispectral bands at time t0 and t1, (c) the iteratively re-weighted multivariate alteration detection (IR-MAD) variates derived from dataset (b); the use of structural information only with image series resulting from the decomposition by the derivative of the morphological profile (DMP) of the panchromatic (d) and pan-sharpened (e) data; finally, the use of spectral and structural information simultaneously (f) and (g) by stacking up (a) and (d), and (b) and (e), respectively. The results show that the SVM performs better with feature vectors based on the simultaneous use of spectral and structural information rather than with those formed by the grey-level information or the DMPs only. Moreover, approach (f) requiring only two panchromatic data as input compete well with approaches (b), (e), and (g), which instead necessitate ten spectral channels as input.

Detection of Complex Salient Regions

The goal of interest point detectors is to find, in an unsupervised way, keypoints easy to extract and at the same time robust to image transformations. We present a novel set of saliency features based on image singularities that takes into account the region content in terms of intensity and local structure. The region complexity is estimated by means of the entropy of the gray-level information; shape information is obtained by measuring the entropy of significant orientations. The regions are located in their representative scale and categorized by their complexity level. Thus, the regions are highly discriminable and less sensitive to confusion and false alarm than the traditional approaches. We compare the novel complex salient regions with the state-of-the-art keypoint detectors. The presented interest points show robustness to a wide set of image transformations and high repeatability as well as allow matching from different camera points of view. Besides, we show the temporal robustness of the novel salient regions in real video sequences, being potentially useful for matching, image retrieval, and object categorization problems.

Development of Volume Conductor and Source Models to Localize Epileptic Foci

There is increasing interest in mapping and source reconstruction from electrocorticoencephalographic (ECoG) grid data and comparison to surface EEG evaluations of epileptic patients. ECoG mapping onto three-dimensional renderings of the individual cortical anatomy derived from magnetic resonance images and computed tomography (CT) is performed after coregistration of anatomical and functional coordinate systems. Source reconstructions from ECoG and EEG are compared using different source models and realistically shaped volume conductor models. Realistically shaped volume conductor models for EEG source reconstruction are a prerequisite for improved localization accuracy. Individual boundary element method (BEM) models derived from MRI represent the "gold standard" and can approximate isotropic homogeneous head compartments and thus give an improved description of the head shape as compared with classical oversimplifying spherical shell models. Anisotropic volume conduction properties of the bone layer or the white matter fibers can be described by the finite element method (FEM); unfortunately, these models require a huge computational effort and are thus not used in daily applications. To avoid this computational effort, head models derived from an averaged MRI dataset can be used. Highly refined models with a large number of nodes and thus better numerical accuracy can be used by this approach, because the setup is performed only once and the decomposed models or precomputed leadfield matrices are saved for later application. Individual image data are not at all needed, if an overlay of the reconstruction results with the anatomy is not desired. With precomputed leadfield matrices and linear interpolation techniques, at least standardized BEM and FEM volume conductor models derived from averaged MRI datasets can achieve the same computational speed as analytical spherical models. The smoothed cortical envelope is used as a realistically shaped single-shell volume conductor model for ECoG source reconstruction, whereas three-compartment BEM-models are required for EEG. The authors describe how to localize ECoG-grid electrode positions and how to segment the cortical surface from coregistered magnetic resonance and CT images. Landmark-based coregistration is performed using common fiducials in both image modalities. Another more promising automatic approach is based on mutual three-dimensional volume gray-level information. The ECoG electrode positions can be retrieved from three-dimensional CT slices manually using cursors in thresholded images with depth information. Alternatively, the smoothed envelope of the cortical surface segmented from the MRI is used to semiautomatically determine the grid electrode positions by marking the four corners and measuring distances along the smoothed surface. With extended source patches for cortically constrained scans and current density reconstructions, results from ECoG and surface EEG data were compared. Single equivalent dipoles were used to explain the EEG far fields, and results were compared with the original current density distributions explaining the ECoG data. The authors studied the performance of spherical and realistically shaped BEM volume conductor models for EEG and ECoG source reconstruction in spherical and nonspherical parts of the head with simulations and measured epileptic spike data. Only small differences between spherical and realistically shaped models were found in the spherical parts of the head, whereas realistically shaped models are superior to spherical approximations in both single-shell ECoG and three-shell EEG cases in the nonspherical parts, such as the temporal lobe areas. The ECoG near field is more complicated to interpret than the surface EEG far field and cannot be explained in general by simple equivalent dipoles. However, from simulations with realistically shaped volume conductor models and cortically constrained source models, the authors studied how the bone and skin layer act as spatial low pass filters that smooth and simplify the surface EEG maps generated by much more complicated-looking source configurations derived from measured ECoG data.

Automatic segmentation and 3D reconstruction of intravascular ultrasound images for a fast preliminar evaluation of vessel pathologies

Intravascular ultrasound (IVUS) imaging is used along with X-ray coronary angiography to detect vessel pathologies. Manual analysis of IVUS images is slow and time-consuming and it is not feasible for clinical purposes. A semi-automated method is proposed to generate 3D reconstructions from IVUS video sequences, so that a fast diagnose can be easily done, quantifying plaque length and severity as well as plaque volume of the vessels under study. The methodology described in this work has four steps: a pre-processing of IVUS images, a segmentation of media–adventitia contour, a detection of intima and plaque and a 3D reconstruction of the vessel. Preprocessing is intended to remove noise from the images without blurring the edges. Segmentation of media–adventitia contour is achieved using active contours (snakes). In particular, we use the gradient vector flow (GVF) as external force for the snakes. The detection of lumen border is obtained taking into account gray-level information of the inner part of the previously detected contours. A knowledge-based approach is used to determine which level of gray corresponds statistically to the different regions of interest: intima, plaque and lumen. The catheter region is automatically discarded. An estimate of plaque type is also given. Finally, 3D reconstruction of all detected regions is made. The suitability of this methodology has been verified for the analysis and visualization of plaque length, stenosis severity, automatic detection of the most problematic regions, calculus of plaque volumes and a preliminary estimation of plaque type obtaining for automatic measures of lumen and vessel area an average error smaller than 1 mm 2 (equivalent aproximately to 10% of the average measure), for calculus of plaque and lumen volume errors smaller than 0.5 mm 3 (equivalent approximately to 20% of the average measure) and for plaque type estimates a mismatch of less than 8% in the analysed frames.

Combined Statistical and Multiscale View on Ultrasonic Liver Images for Characterization

Combined Statistical and Multiscale View on Ultrasonic Liver Images for Characterization

SU‐EE‐A2‐03: Evaluation of Auto‐Segmentation Tools for the Target Definition for the Treatment of Lung Cancer

Purpose: With the advent of more sophisticated image devices in the treatment room, image guided radiotherapy (IGRT) and adaptive radiotherapy (ART) have become distinct possibilities. IGRT and ART techniques in their various stages have been implemented in clinics. One of the ART techniques using the daily acquired CT images involves re‐planning due to the target shape variation during the treatment. Lung cancer volumes of some cases are observed to undergo significant changes where re‐planning is a necessity. To be able to define target efficiently can help the treatment flow significantly. This study evaluates various auto or semi‐auto contouring tools either commercially available or under development for their accuracy and ease of use. Method and Materials: Three methods are included in the study. Two are commercially available (Focal,CMS): auto threshold (of gray level); and auto Segmentation where gray level, the edges and prior shape information are used. The third method is the ITK‐SNAP program that uses a powerful level set(snake) segmentation algorithm to segment anatomical structures in three dimensions. Results: Ten image sets from helical and cone beam CTs are included in the study. The acceptable contours are defined as those with distance to agreement to those drawn by radiation oncologists less than 3 mm. For target volume surrounded by normal lung, the percentage slices of contours that do not need manual adjustment are 41–62%, 23–39%, 62–78% for threshold, auto‐segmentation and SNAP respectively. For cone beam CT, these numbers are approximately 10% lower. SNAP can also be used for target volume with no clear boundary, although the percentage success is much lower. Conclusion: More sophisticated auto‐segmentation tools need to be available routinely with more flexibility for users to adjust algorithm parameters in order for them to be useful for routine clinical ART purposes.

Nonlinear Projection Recovery in Digital Inpainting for Color Image Restoration

This paper concerns with nonuniform sampling and interpolation methods combined with variational models for the solution of a generalized color image inpainting problem and the restoration of digital signals. In particular, we discuss the problem of reconstructing a digital signal/image from very few, sparse, and complete information and from a substantially incomplete information, which will be assumed as the result of a nonlinear distortion. Differently from well known inpainting applications for the recovery of gray images, the proposed techniques apply to color images embedding blanks where only gray level information is given. As a typical and inspiring example, we illustrate the concrete problem of the color restoration of a destroyed art fresco from its few known fragments and some gray picture taken prior to the damage. Numerical implementations are included together with several examples and numerical results to illustrate the proposed method. The numerical experience suggests furthermore that a particular system of coupled Hamilton-Jacobi equations is well-posed.

A coarse-to-fine approach to prostate boundary segmentation in ultrasound images

BackgroundIn this paper a novel method for prostate segmentation in transrectal ultrasound images is presented.MethodsA segmentation procedure consisting of four main stages is proposed. In the first stage, a locally adaptive contrast enhancement method is used to generate a well-contrasted image. In the second stage, this enhanced image is thresholded to extract an area containing the prostate (or large portions of it). Morphological operators are then applied to obtain a point inside of this area. Afterwards, a Kalman estimator is employed to distinguish the boundary from irrelevant parts (usually caused by shadow) and generate a coarsely segmented version of the prostate. In the third stage, dilation and erosion operators are applied to extract outer and inner boundaries from the coarsely estimated version. Consequently, fuzzy membership functions describing regional and gray-level information are employed to selectively enhance the contrast within the prostate region. In the last stage, the prostate boundary is extracted using strong edges obtained from selectively enhanced image and information from the vicinity of the coarse estimation.ResultsA total average similarity of 98.76%(± 0.68) with gold standards was achieved.ConclusionThe proposed approach represents a robust and accurate approach to prostate segmentation.

Computer‐aided detection of breast masses on full field digital mammograms

We are developing a computer-aided detection (CAD) system for breast masses on full field digital mammographic (FFDM) images. To develop a CAD system that is independent of the FFDM manufacturer's proprietary preprocessing methods, we used the raw FFDM image as input and developed a multiresolution preprocessing scheme for image enhancement. A two-stage prescreening method that combines gradient field analysis with gray level information was developed to identify mass candidates on the processed images. The suspicious structure in each identified region was extracted by clustering-based region growing. Morphological and spatial gray-level dependence texture features were extracted for each suspicious object. Stepwise linear discriminant analysis (LDA) with simplex optimization was used to select the most useful features. Finally, rule-based and LDA classifiers were designed to differentiate masses from normal tissues. Two data sets were collected: a mass data set containing 110 cases of two-view mammograms with a total of 220 images, and a no-mass data set containing 90 cases of two-view mammograms with a total of 180 images. All cases were acquired with a GE Senographe 2000D FFDM system. The true locations of the masses were identified by an experienced radiologist. Free-response receiver operating characteristic analysis was used to evaluate the performance of the CAD system. It was found that our CAD system achieved a case-based sensitivity of 70%, 80%, and 90% at 0.72, 1.08, and 1.82 false positive (FP) marks/image on the mass data set. The FP rates on the no-mass data set were 0.85, 1.31, and 2.14 FP marks/image, respectively, at the corresponding sensitivities. This study demonstrated the usefulness of our CAD techniques for automated detection of masses on FFDM images.

A quantile–quantile plot based pattern matching for defect detection

Pattern matching has been used extensively for many machine vision applications such as optical character recognition, face detection, object detection, and defect detection. The normalized cross correlation (NCC) is the most commonly used technique in pattern matching. However, it is computationally intensive, sensitive to environmental changes such as lighting and shifting, and suffers from false alarms for a complicated image that contains partial uniform regions. In this paper, a pattern-matching scheme based on the quantile–quantile plot (Q–Q plot) is proposed for defect detection applications. In a Q–Q plot, the quantiles of an inspection image are plotted against the corresponding quantiles of the template image. The p-value of Chi-square test from the resulting Q–Q plot is then used as the quantitative measure of similarity between two compared images. The quantile representation transforms the 2D gray-level information into the 1D quantile one. It can therefore efficiently reduce the dimensionality of the data, and accelerate the computation. Experimental results have shown that the proposed pattern-matching scheme is computationally fast and is tolerable to minor displacement and process variation. The proposed similarity measure of p-value has excellent discrimination capability to detect subtle defects, compared with the traditional measure of NCC. With a proper normalization of the Q–Q plot, the p-value measure can be tolerable to moderate light changes. Experimental results from assembled PCB (printed circuit board) samples, IC wafers, and liquid crystal display (LCD) panels have shown the efficacy of the proposed pattern-matching scheme for defect detection.

NATURAL SKELETONIZATION: NEW APPROACH FOR THE SKELETONIZATION OF HANDWRITTEN CHARACTERS

In this paper we propose a new algorithm for the skeletonization of handwritten characters. Unlike traditional skeletonization algorithms that relay only on the configuration of a binary image pixel in deciding whether it is deletable or not, Natural Skeletonization (NS) integrates the gray-level information in this process. The underlying principle here, which stems from the elongated properties of the handwritten characters, is that medial pixels of a handwritten stroke are "naturally" darker than its side pixels. NS consists of three steps: (1) the decomposition step; (2) the thinning step; (3) the reconstruction step. The integration of gray-level information is facilitated by the iterative binarization at equally spaced thresholds, which highlights positional differences between the medial and side pixels of a stroke. The advantage of our approach over existing methods is demonstrated by its ability to prevent the "flooding water" and to prevent the boundary noise from developing extraneous branches. One important aspect of the approach is that it relaxes the skeletonization's dependence on the quality and shape of initial binary pattern. The experimental results indicate that the proposed algorithm substantially improves the skeletonization quality compared to experiments with traditional skeletonization methods.

Computerized nipple identification for multiple image analysis in computer‐aided diagnosis

Correlation of information from multiple-view mammograms (e.g., MLO and CC views, bilateral views, or current and prior mammograms) can improve the performance of breast cancer diagnosis by radiologists or by computer. The nipple is a reliable and stable landmark on mammograms for the registration of multiple mammograms. However, accurate identification of nipple location on mammograms is challenging because of the variations in image quality and in the nipple projections, resulting in some nipples being nearly invisible on the mammograms. In this study, we developed a computerized method to automatically identify the nipple location on digitized mammograms. First, the breast boundary was obtained using a gradient-based boundary tracking algorithm, and then the gray level profiles along the inside and outside of the boundary were identified. A geometric convergence analysis was used to limit the nipple search to a region of the breast boundary. A two-stage nipple detection method was developed to identify the nipple location using the gray level information around the nipple, the geometric characteristics of nipple shapes, and the texture features of glandular tissue or ducts which converge toward the nipple. At the first stage, a rule-based method was designed to identify the nipple location by detecting significant changes of intensity along the gray level profiles inside and outside the breast boundary and the changes in the boundary direction. At the second stage, a texture orientation-field analysis was developed to estimate the nipple location based on the convergence of the texture pattern of glandular tissue or ducts towards the nipple. The nipple location was finally determined from the detected nipple candidates by a rule-based confidence analysis. In this study, 377 and 367 randomly selected digitized mammograms were used for training and testing the nipple detection algorithm, respectively. Two experienced radiologists identified the nipple locations which were used as the gold standard. In the training data set, 301 nipples were positively identified and were referred to as visible nipples. Seventy six nipples could not be positively identified and were referred to as invisible nipples. The radiologists provided their estimation of the nipple locations in the latter group for comparison with the computer estimates. The computerized method could detect 89.37% (269/301) of the visible nipples and 69.74% (53/76) of the invisible nipples within 1 cm of the gold standard. In the test data set, 298 and 69 of the nipples were classified as visible and invisible, respectively. 92.28% (275/298) of the visible nipples and 53.62% (37/69) of the invisible nipples were identified within 1 cm of the gold standard. The results demonstrate that the nipple locations on digitized mammograms can be accurately detected if they are visible and can be reasonably estimated if they are invisible. Automated nipple detection will be an important step towards multiple image analysis for CAD.

Modelling of a high resolution digital radiographic system and development of a filtering technique based on wavelet transforms

The first part of this paper describes a geometrical model of an iron cast object with a number of simple defects included in the volume. This model, together with a model of a specific digital X-ray system and procedure, is used to generate a simulated radiograph of the volume. This is then employed in the development of a filtering technique that extracts the grey-level information related to the existence of anomalies in the object. The wavelet-technique is applied in two different ways in order to identify the large-scale of a matrix (i.e. a radiograph). It turns out that both methods are able to remove the geometrical based information from the greyscale in the simulated radiographs. The techniques are also verified with real radiographs taken from a number of welds and iron cast objects with different kinds of defects and defect sizes. The high-resolution digital X-ray system is able to detect defects less than one percentage of the object-thickness in the range of 25–60 mm steel. The developed filtering technique is proven to enhance the detection of defects in objects with geometrical variations and has a potential to reduce time-consuming analysis of radiographs.

An accurate coronary modeling procedure using 2D calibrated projections based on extracted 3D vessel centerlines

This paper describes a method for the three-dimensional (3D) reconstruction of coronary arterial segments from three projection boundaries. The method is based on the 3D generalized cylinder (GC) model with elliptic cross-sections and mainly consists of a model-based estimation of the vessel boundaries, precise computation of the ellipse parameters of arterial cross-sections from the estimated boundaries, and obtaining the 3D structures by linking consecutive cross-sections. Synthetic and real experiments showed that the method is robust to degradation sources and consequently superior to the two-view method because it uses boundary instead of gray-level information.

“Do Not (Mis-)Adjust Your Set”: Maintaining Specimen Detail in the Video Microscope

This chapter discusses the practical aspects of adjusting a video camera and a monitor to prevent the loss of specimen image gray-level information. To maximize the amount of specimen information that is displayed on the video monitor or recorded in electronic format, all components of the video microscopy imaging system must be optimized: microscope, camera coupling to the microscope, camera, image processing system, and monitor. Failure to properly control any of these links in the integrated imaging system inevitably leads to the loss of specimen detail.

Image analysis system for detection facial paralysis

Evaluation of facial nerve paresis depends on visual assessment and naturally differs from examiner to examiner. An objective measurement instrument is presented. Facial features are automatically localized by a parametric face model in videos of a face during relaxation and exercises. Gray-level information is analyzed by a special steerable filter and used to identify symmetries. The computer system was tested in 19 individuals. Automatic localization of facial features such as the upper arc of the head and ears was correct in 95%, the eyes in 82%, and the mouth in 73%. Lid paresis was correctly recognized in seven of ten (70%) and oral paresis in 10 of 12 (83%) cases. Unaffected eyelid movements were identified in eight of nine (89%) and healthy oral regions in all seven (100%) cases. The computer system presented is able to automatically localize facial features and to identify facial nerve paresis. It is a considerable step toward automatic and objective grading of facial nerve paresis.

Endometrium characterization in connection with uterine contraction

The features of the endometrium are considered important factors in infertility treatment. This paper presents a new approach for estimating the endometrium based on a multiple threshold technique and a general mathematical method. The approach exploits gray-level information to organize and constrain the interpretation of the motion object by a recursive model. The recursive model is employed for threshold optimization of motion information segmentation. This is implemented by repeatedly shrinking the dimension of the maximum potential range of motion with an adaptive threshold throughout the whole process. The values to be estimated comprise motion frequency and thickness. The former is determined by change detection of the number of frame based changes in a sequence of images over a given time period with an appropriate sampling frequency. This can be achieved by a double-increment-of-sampling-frequency mechanism. The latter is the result of the motion object's dimension divided by its extended length. Experimental results indicate that it is possible to accurately estimate the thickness and motion frequency of the endometrium with the proposed algorithm.

Accurate Building Structure Recovery from High Resolution Aerial Imagery

This paper focuses on the problem of automatic extraction and modeling of buildings from stereoscopic pairs of high resolution aerial images covering urban areas. An increasing number of applications require accurate and up-to-date cartographic and 3-D data.We introduce a set of algorithms based on an accurate, reliable, and discontinuity-preserving digital elevation model (DEM) computation. This method involves a gradient correlation, contour-adaptive windows with a geodesic weighting, a multi- resolution coarse to fine scheme, and a two-way filtering symmetrical validation. The DEMs are then segmented and each region is classified either as ground, vegetation, or building. The regions of interest, i.e., the buildings, are then individually processed, and a two-step modeling process is applied. First, building boundaries are detected and linearized: segments are extracted from the images, height and gray-level information computed in their neighborhoods is added to their description, and segments are grouped taking into account some constraints on neighborhood compatibility. Second, we develop an original method for the estimation of roof surface planes based on a Bayesian approach. This method directly provides the number of planes and their equations from the 3-D dataset of the regions of interest.The method involves thresholds, most of them tuned with respect to the physical characteristics of the scene, and only two parameters have to be adjusted. Results are shown and discussed on four different stereo pairs.