We present a method for the automatic localization of facial landmarks that integrates nonrigid deformation with the ability to handle missing points. The algorithm generates sets of candidate locations from feature detectors and performs combinatorial search constrained by a flexible shape model. A key assumption of our approach is that for some landmarks there might not be an accurate candidate in the input set. This is tackled by detecting partial subsets of landmarks and inferring those that are missing, so that the probability of the flexible model is maximized. The ability of the model to work with incomplete information makes it possible to limit the number of candidates that need to be retained, drastically reducing the number of combinations to be tested with respect to the alternative of trying to always detect the complete set of landmarks. We demonstrate the accuracy of the proposed method in the face recognition grand challenge database, where we obtain average errors of approximately 3.5 mm when targeting 14 prominent facial landmarks. For the majority of these our method produces the most accurate results reported to date in this database. Handling of occlusions and surfaces with missing parts is demonstrated with tests on the Bosphorus database, where we achieve an overall error of 4.81 and 4.25 mm for data with and without occlusions, respectively. To investigate potential limits in the accuracy that could be reached, we also report experiments on a database of 144 facial scans acquired in the context of clinical research, with manual annotations performed by experts, where we obtain an overall error of 2.3 mm, with averages per landmark below 3.4 mm for all 14 targeted points and within 2 mm for half of them. The coordinates of automatically located landmarks are made available on-line.

This paper evaluates the performance of state of the art colour invariants for the purposes of local image feature detection. We adapt the Harris-Laplace detector for colour invariance and test it under general image distortions. A second investigation examines the correlation between the colour invariants where the number of correctly detected unique points are analysed. This paper aims to answer if colour invariants should be used for feature detection purposes, and if they could be jointly used by feature fusion techniques to augment the performance of intensity-based detectors.

In this paper, we investigate the segmentation of closed contours in subcellular data using a framework that primarily combines the pairwise affinity grouping principles with a graph partitioning contour searching approach. One salient problem that precluded the application of these methods to large scale segmentation problems is the onerous computational complexity required to generate comprehensive representations that include all pairwise relationships between all pixels in the input data. To compensate for this problem, a practical solution is to reduce the complexity of the input data by applying an over-segmentation technique prior to the application of the computationally demanding strands of the segmentation process. This approach opens the opportunity to build specific shape and intensity models that can be successfully employed to extract the salient structures in the input image which are further processed to identify the cycles in an undirected graph. The proposed framework has been applied to the segmentation of mitochondria membranes in electron microscopy data which are characterized by low contrast and low signal-to-noise ratio. The algorithm has been quantitatively evaluated using two datasets where the segmentation results have been compared with the corresponding manual annotations. The performance of the proposed algorithm has been measured using standard metrics, such as precision and recall, and the experimental results indicate a high level of segmentation accuracy.

Histogram transformation defines a class of image processing operations that are widely applied in the implementation of data normalization algorithms. In this paper, we present a new variational approach for image enhancement that is constructed to alleviate the intensity saturation effects that are introduced by standard contrast enhancement (CE) methods based on histogram equalization. In this paper, we initially apply total variation (TV) minimization with a L(1) fidelity term to decompose the input image with respect to cartoon and texture components. Contrary to previous papers that rely solely on the information encompassed in the distribution of the intensity information, in this paper, the texture information is also employed to emphasize the contribution of the local textural features in the CE process. This is achieved by implementing a nonlinear histogram warping CE strategy that is able to maximize the information content in the transformed image. Our experimental study addresses the CE of a wide variety of image data and comparative evaluations are provided to illustrate that our method produces better results than conventional CE strategies.

The aim of this paper is to detail the development of a novel tracking framework that is able to extract the cell motility indicators and to determine the cellular division (mitosis) events in large time-lapse phase-contrast image sequences. To address the challenges induced by nonstructured (random) motion, cellular agglomeration, and cellular mitosis, the process of automatic (unsupervised) cell tracking is carried out in a sequential manner, where the interframe cell association is achieved by assessing the variation in the local cellular structures in consecutive frames of the image sequence. In our study, a strong emphasis has been placed on the robust use of the topological information in the cellular tracking process and in the development of targeted pattern recognition techniques that were designed to redress the problems caused by segmentation errors, and to precisely identify mitosis using a backward (reversed) tracking strategy. The proposed algorithm has been evaluated on dense phase-contrast cellular data and the experimental results indicate that the proposed algorithm is able to accurately track epithelial and endothelial cells in time-lapse image sequences that are characterized by low contrast and high level of noise. Our algorithm achieved 86.10% overall tracking accuracy and 90.12% mitosis detection accuracy.

A common approach to model-based segmentation is to assume a top-down modelling strategy. However, this is not feasible for complex 3D +time structures, such as the cardiac left ventricle, due to increased training requirements, aligning difficulties and local minima in resulting models. As our main contribution, we present an alternate bottom-up modelling approach. By combining the variation captured in multiple dimensionally-targeted models at segmentation-time we create a scalable segmentation framework that does not suffer from the "curse of dimensionality." Our second contribution involves a flexible contour coupling technique that allows our segmentation method to adapt to unseen contour configurations outside the training set. This is used to identify the endo- and epicardium contours of the left ventricle by coupling them at segmentation-time, instead of at model-time. We apply our approach to 33 3D +time cardiac MRI datasets and perform comprehensive evaluation against several state-of-the-art works. Quantitative evaluation illustrates that our method requires significantly less training than state-of-the-art model-based methods, while maintaining or improving segmentation accuracy.

With the proliferation of modern microscopy imaging technologies the amount of data that has to be analysed by biologists is constantly increasing and as a result the development of automatic approaches that are able to track cellular structures in time-lapse images has become an important field of research. The aim of this paper is to detail the development of a novel tracking framework that is designed to extract the cell motility indicators in phase-contrast image sequences. To address issues that are caused by non-structured (random) motion and cellular agglomeration, cell tracking is formulated as a sequential process where the inter-frame cell association is achieved by assessing the variation in the local structures contained in consecutive frames of the image sequence. We have evaluated the proposed algorithm on dense phase contrast cellular data and the reported results indicate that the developed algorithm is able to accurately track Madin-Darby Canine Kidney (MDCK) Epithelial Cells in image data that is characterised by low contrast and high level of noise.

The advent of new hyperspectral imaging modalities made possible the implementation of flexible machine vision systems that can be designed to solve a variety of industrial tasks such as automatic material sorting. However the design of robust machine vision systems is far from a trivial task as several issues including mechanical design, development of an appropriate illumination set-up, optimal interfacing between the sensing and optical equipment with the computer vision component have to be properly addressed in order to accommodate all challenges that are encountered in a typical industrial environment. In this paper we present a novel methodology to automate the recycling process of non-ferrous metal Waste from Electric and Electronic Equipment (WEEE) where a particular emphasis is placed on the design choices that were made in the development of the proposed waste sorting system. The developed machine vision system has been subjected to a thorough robustness evaluation and the reported experimental results indicate that the proposed solution can be used to replace the manual procedure that is currently used in WEEE recycling plants.

In this paper we present a parameter optimisation procedure that is designed to automatically initialise the number of clusters and the initial colour prototypes required by data space partitioning techniques. The proposed optimisation approach involves a colour saliency measure used in conjunction with a SOM classification procedure. For evaluation purposes, we have integrated the proposed initialisation technique in an unsupervised colour segmentation scheme based on K-Means clustering and the evaluation has been carried out in the context of the unsupervised segmentation of natural images.

Hyperspectral data allows the construction of more elaborate models to sample the properties of the nonferrous materials than the standard RGB color representation. In this paper, the nonferrous waste materials are studied as they cannot be sorted by classical procedures due to their color, weight and shape similarities. The experimental results presented in this paper reveal that factors such as the various levels of oxidization of the waste materials and the slight differences in their chemical composition preclude the use of the spectral features in a simplistic manner for robust material classification. To address these problems, the proposed FUSSER (fuzzy spectral and spatial classifier) algorithm detailed in this paper merges the spectral and spatial features to obtain a combined feature vector that is able to better sample the properties of the nonferrous materials than the single pixel spectral features when applied to the construction of multivariate Gaussian distributions. This approach allows the implementation of statistical region merging techniques in order to increase the performance of the classification process. To achieve an efficient implementation, the dimensionality of the hyperspectral data is reduced by constructing bio-inspired spectral fuzzy sets that minimize the amount of redundant information contained in adjacent hyperspectral bands. The experimental results indicate that the proposed algorithm increased the overall classification rate from 44% using RGB data up to 98% when the spectral-spatial features are used for nonferrous material classification.

This paper presents the development of an unsupervised image segmentation framework (referred to as CTex) that is based on the adaptive inclusion of color and texture in the process of data partition. An important contribution of this work consists of a new formulation for the extraction of color features that evaluates the input image in a multispace color representation. To achieve this, we have used the opponent characteristics of the RGB and YIQ color spaces where the key component was the inclusion of the Self Organizing Map (SOM) network in the computation of the dominant colors and estimation of the optimal number of clusters in the image. The texture features are computed using a multichannel texture decomposition scheme based on Gabor filtering. The major contribution of this work resides in the adaptive integration of the color and texture features in a compound mathematical descriptor with the aim of identifying the homogenous regions in the image. This integration is performed by a novel adaptive clustering algorithm that enforces the spatial continuity during the data assignment process. A comprehensive qualitative and quantitative performance evaluation has been carried out and the experimental results indicate that the proposed technique is accurate in capturing the color and texture characteristics when applied to complex natural images.

Modern medical imaging modalities provide large amounts of information in both the spatial and temporal domains and the incorporation of this information in a coherent algorithmic framework is a significant challenge. In this paper, we present a novel and intuitive approach to combine 3-D spatial and temporal (3-D + time) magnetic resonance imaging (MRI) data in an integrated segmentation algorithm to extract the myocardium of the left ventricle. A novel level-set segmentation process is developed that simultaneously delineates and tracks the boundaries of the left ventricle muscle. By encoding prior knowledge about cardiac temporal evolution in a parametric framework, an expectation-maximization algorithm optimally tracks the myocardial deformation over the cardiac cycle. The expectation step deforms the level-set function while the maximization step updates the prior temporal model parameters to perform the segmentation in a nonrigid sense.

Generic camera calibration is a non-parametric calibration technique that is applicable to any type of vision sensor. However, the standard generic calibration method was developed with the goal of generality and it is therefore sub-optimal for the common case of cameras with a single centre of projection (e.g. pinhole, fisheye, hyperboloidal catadioptric). This paper proposes novel improvements to the standard generic calibration method for central cameras that reduce its complexity, and improve its accuracy and robustness. Improvements are achieved by taking advantage of the geometric constraints resulting from a single centre of projection. Input data for the algorithm is acquired using active grids, the performance of which is characterised. A new linear estimation stage to the generic algorithm is proposed incorporating classical pinhole calibration techniques, and it is shown to be significantly more accurate than the linear estimation stage of the standard method. A linear method for pose estimation is also proposed and evaluated against the existing polynomial method. Distortion correction and motion reconstruction experiments are conducted with real data for a hyperboloidal catadioptric sensor for both the standard and proposed methods. Results show the accuracy and robustness of the proposed method to be superior to those of the standard method.

We present a preprocessing and segmentation scheme designed to address the particular difficulties encountered in the analysis of magnetic resonance cholangiopancreatography (MRCP) data, as a precursor to the application of computer assisted diagnosis (CAD) techniques. MRCP generates noisy, low resolution, non-isometric data which often exhibits significant greylevel inhomogeneities. This combination of characteristics results in data volumes in which reliable segmentation and analysis are difficult to achieve. In this paper we describe a data processing approach developed to overcome these difficulties and allow for the effective application of automated CAD procedures in the analysis of the biliary tree and pancreatic duct in MRCP examinations.

The ability to manipulate previously unseen objects under visual control is one of the key tasks in the successful implementation of robotic, automated assembly and adaptive material handling systems. The automation of such complex industrial environments will require the development of machine vision systems that are highly adaptable and capable of dealing with a wide range of variable products. An important group of applications of this type is found in the automated packing and nesting of arbitrary shapes. The aim of this work has been to produce an efficient packing strategy that is flexible enough for a wide variety of industrial uses and which can be implemented using fast moderately priced hardware. A systems approach, as distinct from a purely algorithmic one, has been deliberately adopted since the work is concerned with industrial vision applications in which significant problem constraints exist. This paper also outlines the background to this research, and reviews a selection of industrial packing applications. The packing procedure that has been devised, consists of two major components. The first is a geometric packing technique that is based on morphological image processing operations. This is used in conjunction with a prolog based heuristic packing procedure. Some of the factors considered at the heuristic level include shape ordering and shape orientation, both of which must be carried out prior to the implementation of the geometric packer. The heuristic procedures deal with problem constraints that are specific to a given application.