Bob Fisher | University of Edinburgh (original) (raw)

Papers by Bob Fisher

We propose a system for describing skin lesions images based on a human perception model. Pigment... more We propose a system for describing skin lesions images based on a human perception model. Pigmented skinlesions including melanoma and other types of skin cancer as well as non-malignant lesions are used. Works onclassification of skin lesions already exist but they mainly concentrate on melanoma. The novelty of our work isthat our system gives to skin lesion images a semantic label in a manner similar to humans. This work consists of two parts: first we capture they way users perceive each lesion, second we train a machine learning system thatsimulates how people describe images. For the first part, we choose 5 attributes: colour (light to dark), colouruniformity (uniform to non-uniform), symmetry (symmetric to non-symmetric), border (regular to irregular),texture (smooth to rough). Using a web based form we asked people to pick a value of each attribute for eachlesion. In the second part, we extract 93 features from each lesions and we trained a machine learning algorithmusing such features as input and the values of the human attributes as output. Results are quite promising,especially for the colour related attributes, where our system classifies over 80% of the lesions into the samesemantic classes as humans.

The ability to construct CAD or other object models from edge and range data has a fundamental me... more The ability to construct CAD or other object models from edge and range data has a fundamental meaning in building a recognition and positioning system. While the problem of model fitting has been successfully addressed, the problem of efficient high accuracy and stability of the fitting is still an open problem. In the past researchers have used approximate distance functions rather than the real Euclidean distance because of computational efficiency. We now feel that machine speeds are sufficient to ask whether it is worth considering Euclidean fitting again. This paper address the problem of estimation of elliptical cylinder and cone surfaces to 3D data by a constrained Euclidean fitting. We study and compare the performance of various distance functions in terms of correctness, robustness and pose invariance, and present our results improving known fitting methods by closed form expressions of the real Euclidean distance.

The paper addresses two problems related to 3D camera calibration using a single mono-plane calib... more The paper addresses two problems related to 3D camera calibration using a single mono-plane calibration target with circular control marks. The first problem is how to compute accurately the locations of the features (ellipses) in images of the target. Since the structure of the control marks is known beforehand, we propose to use a shape-specific searching technique to find the optimal locations of the features. Our experiments have shown this technique generates more accurate feature locations than the state-of-the-art ellipse extraction methods. The second problem is how to refine the control mark locations with unknown manufacturing errors. We demonstrate in a case study, where the control marks are laser printed on a A4 paper, that the manufacturing errors of the control marks can be compensated to a good extent so that the remaining calibration errors are reduced significantly.

Noblesse Workshop on Non-Linear Model Based Image Analysis, 1998

In this paper we present a novel representation for arbitrary surfaces that enables local corresp... more In this paper we present a novel representation for arbitrary surfaces that enables local correspondences to be determined. We then describe how these local correspondences can be used to search for the transformation that best aligns all of surface data. If this transformation is found to align a signi cant proportion of the surface data then the surfaces are said to have a correspondence.

Advances in Soft Computing, 1999

Medical Imaging 2011: Image Perception, Observer Performance, and Technology Assessment, 2011

In this work evidence is presented supporting the hypothesis that observers tend to evaluate very... more In this work evidence is presented supporting the hypothesis that observers tend to evaluate very differently the same properties of given skin-lesion images. Results from previous experiments have been compared to new ones obtained where we gave additional prototypical visual cues to the users during their evaluation trials. Each property (colour, colour uniformity, asymmetry, border regularity, roughness of texture) had to be evaluated on a 0-10 range, with both linguistic descriptors and visual references at each end and in the middle (e.g. light/medium/dark for colour). A set of 22 images covering different clinical diagnoses has been used in the comparison with previous results. Statistical testing showed that only for a few test images the inclusion of the visual anchors reduced the variability of the grading for some of the properties. Despite such reduction, though, the average variance of each property still remains high even after the inclusion of the visual anchors. When considering each property, the average variance significantly changed for the roughness of texture, where the visual references caused an increase in the variability. With these results we can conclude that the variance of the answers observed in the previous experiments was not due to the lack of a standard definition of the extrema of the scale, but rather to a high variability in the way observers perceive and understand skin-lesion images.

Pattern Recognition, 2015

Classification of data is difficult if the data is imbalanced and classes are overlapping. In rec... more Classification of data is difficult if the data is imbalanced and classes are overlapping. In recent years, more research has started to focus on classification of imbalanced data since real world data is often skewed. Traditional methods are more successful with classifying the class that has the most samples (majority class) compared to the other classes (minority classes). For the classification of imbalanced data sets, different methods are available, although each has some advantages and shortcomings. In this study, we propose a new hierarchical decomposition method for imbalanced data sets which is different from previously proposed solutions to the class imbalance problem. Additionally, it does not require any data pre-processing step as many other solutions need. The new method is based on clustering and outlier detection. The hierarchy is constructed using the similarity of labeled data subsets at each level of the hierarchy with different levels being built by different data and feature subsets. Clustering is used to partition the data while outlier detection is utilized to detect minority class samples. The comparison of the proposed method with state of art the methods using 20 public imbalanced data sets and 181 synthetic data sets showed that the proposed method's classification performance is better than the state of art methods. It is especially successful if the minority class is sparser than the majority class. It has accurate performance even when classes have sub-varieties and minority and majority classes are overlapping. Moreover, its performance is also good when the class imbalance ratio is low, i.e. classes are more imbalanced.

Procedings of the British Machine Vision Conference 2013, 2013

ABSTRACT We address the analysis of fish trajectories in unconstrained underwater videos to help ... more ABSTRACT We address the analysis of fish trajectories in unconstrained underwater videos to help marine biologist to detect new/rare fish behaviours and to detect environmental changes which can be observed from the abnormal behaviour of fish. The fish trajectories are separated into normal and abnormal classes which indicate the common behaviour of fish and the behaviours that are rare/ unusual respectively. The proposed solution is based on a novel type of hierarchical classifier which builds the tree using clustered and labelled data based on similarity of data while using different feature sets at different levels of hierarchy. The paper presents a new method for fish trajectory analysis which has better performance compared to state-of-the-art techniques while the results are significant considering the challenges of underwater environments, low video quality, erratic movement of fish and highly imbalanced trajectory data that we used. Moreover, the proposed method is also powerful enough to classify highly imbalanced real-world datasets.

International Journal of Computer Vision, 2012

This special issue discusses new results, technologies and applications related to the capture, r... more This special issue discusses new results, technologies and applications related to the capture, representation, processing, storage, transmission and visualization of 3D geometric and photometric models.

The ability to construct CAD or other object models from edge and range data has a fundamental me... more The ability to construct CAD or other object models from edge and range data has a fundamental meaning in building a recognition and positioning system. While the problem of model fitting has been successfully addressed, the problem of efficient high accuracy and stability of the fitting is still an open problem. In the past researchers have used approximate distance functions rather than the real Euclidean distance because of computational efficiency. We now feel that machine speeds are sufficient to ask whether it is worth considering Euclidean fitting again. This paper address the problem of estimation of elliptical cylinder and cone surfaces to 3D data by a constrained Euclidean fitting. We study and compare the performance of various distance functions in terms of correctness, robustness and pose invariance, and present our results improving known fitting methods by closed form expressions of the real Euclidean distance.

ABSTRACT Long-term monitoring of the underwater environment is still labour intensive work. Using... more ABSTRACT Long-term monitoring of the underwater environment is still labour intensive work. Using underwater surveillance cameras to monitor this environment has the potential advantage to make the task become less labour intensive. Also, the obtained data can be stored making the research reproducible. In this work, a system to analyse long-term underwater camera footage (more than 3 years of 12 hours a day underwater camera footage from 10 cameras) is described. This system uses video processing software to detect and recognise fish species. This footage is processed on supercomputers, which allow marine biologists to request automatic processing on these videos and afterwards analyse the results using a web-interface that allows them to display counts of fish species in the camera footage.

We explore the potential of variance matrices to represent not just statistical error on object p... more We explore the potential of variance matrices to represent not just statistical error on object pose estimates but also partially constrained degrees of freedom. Using an iterated extended Kalman filter as an estimation tool, we generate, combine and predict partially constrained pose estimates from 3D range data. We find that partial constraints on the translation component of pose which occur frequently in practice are handled well by the method. One key advantage of the method is that it allows simultaneous representation of both lack of knowledge, weak constraints, a priori position constraints and statistical error in a framework that allows incremental reasoning.

This paper outlines the automatic construction of video processing solutions using multiple softw... more This paper outlines the automatic construction of video processing solutions using multiple software components as opposed to traditional monolithic approaches undertaken by image processing experts. A combined top-down and bottomup methodology was adopted for the derivation of a suitable level of granularity for a subset of image processing components that implement video classification, object detection, counting and tracking tasks. 90% of these components are generic and could be applied to any video processing task, indicating a high level of reusability for a spectrum of video analyses. Domainspecific video analysis approaches (that exploit combinations of the above components) are built by using an automatic workflow composition module that relies on decomposition-based planning and ontologies. Evaluation on a set of ecological videos indicate that the proposed approach is faster and more flexible to adapt to changes in domain descriptions than specialized components written from scratch by image processing experts.

In this paper, a novel model of object-based visual attention extending Duncan's Integrated Compe... more In this paper, a novel model of object-based visual attention extending Duncan's Integrated Competition Hypothesis [24] is presented. In contrast to the attention mechanisms used in most previous machine vision systems which drive attention based on the spatial location hypothesis, the mechanisms which direct visual attention in our system are object-driven as well as feature-driven. The competition to gain visual attention occurs not only within an object but also between objects. For this purpose, two new mechanisms in the proposed model are described and analyzed in detail. The first mechanism computes the visual salience of objects and groupings; the second one implements the hierarchical selectivity of attentional shifts. The results of the new approach on synthetic and natural images are reported.

We present an algorithm for determining the next best position of a range sensor in 3D space for ... more We present an algorithm for determining the next best position of a range sensor in 3D space for incrementally recovering an indoor scene. The method works in ve dimensions: the sensor navigates inside the scene and can be placed at any 3D position and oriented by a pan-tilt head. The method is based on a mixed exhaustive search and hill climbing optimisation, and outputs the next position in usable time. Results are shown with a simulated mobile range sensor navigating in CAD models of environments (closed rooms).

This paper shows that adding 3D depth information to RGB colour images improves segmentation of p... more This paper shows that adding 3D depth information to RGB colour images improves segmentation of pigmented and non-pigmented skin lesion. A regionbased active contour segmentation approach using a statistical model based on the level-set framework is presented. We consider what kinds of properties (e.g., colour, depth, texture) are most discriminative. The experiments show that our proposed method integrating chromatic and geometric information produces segmentation results for pigmented lesions close to dermatologists and more consistent and accurate results for non-pigmented lesions.

We propose a system for describing skin lesions images based on a human perception model. Pigment... more We propose a system for describing skin lesions images based on a human perception model. Pigmented skinlesions including melanoma and other types of skin cancer as well as non-malignant lesions are used. Works onclassification of skin lesions already exist but they mainly concentrate on melanoma. The novelty of our work isthat our system gives to skin lesion images a semantic label in a manner similar to humans. This work consists of two parts: first we capture they way users perceive each lesion, second we train a machine learning system thatsimulates how people describe images. For the first part, we choose 5 attributes: colour (light to dark), colouruniformity (uniform to non-uniform), symmetry (symmetric to non-symmetric), border (regular to irregular),texture (smooth to rough). Using a web based form we asked people to pick a value of each attribute for eachlesion. In the second part, we extract 93 features from each lesions and we trained a machine learning algorithmusing such features as input and the values of the human attributes as output. Results are quite promising,especially for the colour related attributes, where our system classifies over 80% of the lesions into the samesemantic classes as humans.

The ability to construct CAD or other object models from edge and range data has a fundamental me... more The ability to construct CAD or other object models from edge and range data has a fundamental meaning in building a recognition and positioning system. While the problem of model fitting has been successfully addressed, the problem of efficient high accuracy and stability of the fitting is still an open problem. In the past researchers have used approximate distance functions rather than the real Euclidean distance because of computational efficiency. We now feel that machine speeds are sufficient to ask whether it is worth considering Euclidean fitting again. This paper address the problem of estimation of elliptical cylinder and cone surfaces to 3D data by a constrained Euclidean fitting. We study and compare the performance of various distance functions in terms of correctness, robustness and pose invariance, and present our results improving known fitting methods by closed form expressions of the real Euclidean distance.

The paper addresses two problems related to 3D camera calibration using a single mono-plane calib... more The paper addresses two problems related to 3D camera calibration using a single mono-plane calibration target with circular control marks. The first problem is how to compute accurately the locations of the features (ellipses) in images of the target. Since the structure of the control marks is known beforehand, we propose to use a shape-specific searching technique to find the optimal locations of the features. Our experiments have shown this technique generates more accurate feature locations than the state-of-the-art ellipse extraction methods. The second problem is how to refine the control mark locations with unknown manufacturing errors. We demonstrate in a case study, where the control marks are laser printed on a A4 paper, that the manufacturing errors of the control marks can be compensated to a good extent so that the remaining calibration errors are reduced significantly.

Noblesse Workshop on Non-Linear Model Based Image Analysis, 1998

In this paper we present a novel representation for arbitrary surfaces that enables local corresp... more In this paper we present a novel representation for arbitrary surfaces that enables local correspondences to be determined. We then describe how these local correspondences can be used to search for the transformation that best aligns all of surface data. If this transformation is found to align a signi cant proportion of the surface data then the surfaces are said to have a correspondence.

Advances in Soft Computing, 1999

Medical Imaging 2011: Image Perception, Observer Performance, and Technology Assessment, 2011

In this work evidence is presented supporting the hypothesis that observers tend to evaluate very... more In this work evidence is presented supporting the hypothesis that observers tend to evaluate very differently the same properties of given skin-lesion images. Results from previous experiments have been compared to new ones obtained where we gave additional prototypical visual cues to the users during their evaluation trials. Each property (colour, colour uniformity, asymmetry, border regularity, roughness of texture) had to be evaluated on a 0-10 range, with both linguistic descriptors and visual references at each end and in the middle (e.g. light/medium/dark for colour). A set of 22 images covering different clinical diagnoses has been used in the comparison with previous results. Statistical testing showed that only for a few test images the inclusion of the visual anchors reduced the variability of the grading for some of the properties. Despite such reduction, though, the average variance of each property still remains high even after the inclusion of the visual anchors. When considering each property, the average variance significantly changed for the roughness of texture, where the visual references caused an increase in the variability. With these results we can conclude that the variance of the answers observed in the previous experiments was not due to the lack of a standard definition of the extrema of the scale, but rather to a high variability in the way observers perceive and understand skin-lesion images.

Pattern Recognition, 2015

Classification of data is difficult if the data is imbalanced and classes are overlapping. In rec... more Classification of data is difficult if the data is imbalanced and classes are overlapping. In recent years, more research has started to focus on classification of imbalanced data since real world data is often skewed. Traditional methods are more successful with classifying the class that has the most samples (majority class) compared to the other classes (minority classes). For the classification of imbalanced data sets, different methods are available, although each has some advantages and shortcomings. In this study, we propose a new hierarchical decomposition method for imbalanced data sets which is different from previously proposed solutions to the class imbalance problem. Additionally, it does not require any data pre-processing step as many other solutions need. The new method is based on clustering and outlier detection. The hierarchy is constructed using the similarity of labeled data subsets at each level of the hierarchy with different levels being built by different data and feature subsets. Clustering is used to partition the data while outlier detection is utilized to detect minority class samples. The comparison of the proposed method with state of art the methods using 20 public imbalanced data sets and 181 synthetic data sets showed that the proposed method's classification performance is better than the state of art methods. It is especially successful if the minority class is sparser than the majority class. It has accurate performance even when classes have sub-varieties and minority and majority classes are overlapping. Moreover, its performance is also good when the class imbalance ratio is low, i.e. classes are more imbalanced.

Procedings of the British Machine Vision Conference 2013, 2013

ABSTRACT We address the analysis of fish trajectories in unconstrained underwater videos to help ... more ABSTRACT We address the analysis of fish trajectories in unconstrained underwater videos to help marine biologist to detect new/rare fish behaviours and to detect environmental changes which can be observed from the abnormal behaviour of fish. The fish trajectories are separated into normal and abnormal classes which indicate the common behaviour of fish and the behaviours that are rare/ unusual respectively. The proposed solution is based on a novel type of hierarchical classifier which builds the tree using clustered and labelled data based on similarity of data while using different feature sets at different levels of hierarchy. The paper presents a new method for fish trajectory analysis which has better performance compared to state-of-the-art techniques while the results are significant considering the challenges of underwater environments, low video quality, erratic movement of fish and highly imbalanced trajectory data that we used. Moreover, the proposed method is also powerful enough to classify highly imbalanced real-world datasets.

International Journal of Computer Vision, 2012

This special issue discusses new results, technologies and applications related to the capture, r... more This special issue discusses new results, technologies and applications related to the capture, representation, processing, storage, transmission and visualization of 3D geometric and photometric models.

The ability to construct CAD or other object models from edge and range data has a fundamental me... more The ability to construct CAD or other object models from edge and range data has a fundamental meaning in building a recognition and positioning system. While the problem of model fitting has been successfully addressed, the problem of efficient high accuracy and stability of the fitting is still an open problem. In the past researchers have used approximate distance functions rather than the real Euclidean distance because of computational efficiency. We now feel that machine speeds are sufficient to ask whether it is worth considering Euclidean fitting again. This paper address the problem of estimation of elliptical cylinder and cone surfaces to 3D data by a constrained Euclidean fitting. We study and compare the performance of various distance functions in terms of correctness, robustness and pose invariance, and present our results improving known fitting methods by closed form expressions of the real Euclidean distance.

ABSTRACT Long-term monitoring of the underwater environment is still labour intensive work. Using... more ABSTRACT Long-term monitoring of the underwater environment is still labour intensive work. Using underwater surveillance cameras to monitor this environment has the potential advantage to make the task become less labour intensive. Also, the obtained data can be stored making the research reproducible. In this work, a system to analyse long-term underwater camera footage (more than 3 years of 12 hours a day underwater camera footage from 10 cameras) is described. This system uses video processing software to detect and recognise fish species. This footage is processed on supercomputers, which allow marine biologists to request automatic processing on these videos and afterwards analyse the results using a web-interface that allows them to display counts of fish species in the camera footage.

We explore the potential of variance matrices to represent not just statistical error on object p... more We explore the potential of variance matrices to represent not just statistical error on object pose estimates but also partially constrained degrees of freedom. Using an iterated extended Kalman filter as an estimation tool, we generate, combine and predict partially constrained pose estimates from 3D range data. We find that partial constraints on the translation component of pose which occur frequently in practice are handled well by the method. One key advantage of the method is that it allows simultaneous representation of both lack of knowledge, weak constraints, a priori position constraints and statistical error in a framework that allows incremental reasoning.

This paper outlines the automatic construction of video processing solutions using multiple softw... more This paper outlines the automatic construction of video processing solutions using multiple software components as opposed to traditional monolithic approaches undertaken by image processing experts. A combined top-down and bottomup methodology was adopted for the derivation of a suitable level of granularity for a subset of image processing components that implement video classification, object detection, counting and tracking tasks. 90% of these components are generic and could be applied to any video processing task, indicating a high level of reusability for a spectrum of video analyses. Domainspecific video analysis approaches (that exploit combinations of the above components) are built by using an automatic workflow composition module that relies on decomposition-based planning and ontologies. Evaluation on a set of ecological videos indicate that the proposed approach is faster and more flexible to adapt to changes in domain descriptions than specialized components written from scratch by image processing experts.

In this paper, a novel model of object-based visual attention extending Duncan's Integrated Compe... more In this paper, a novel model of object-based visual attention extending Duncan's Integrated Competition Hypothesis [24] is presented. In contrast to the attention mechanisms used in most previous machine vision systems which drive attention based on the spatial location hypothesis, the mechanisms which direct visual attention in our system are object-driven as well as feature-driven. The competition to gain visual attention occurs not only within an object but also between objects. For this purpose, two new mechanisms in the proposed model are described and analyzed in detail. The first mechanism computes the visual salience of objects and groupings; the second one implements the hierarchical selectivity of attentional shifts. The results of the new approach on synthetic and natural images are reported.

We present an algorithm for determining the next best position of a range sensor in 3D space for ... more We present an algorithm for determining the next best position of a range sensor in 3D space for incrementally recovering an indoor scene. The method works in ve dimensions: the sensor navigates inside the scene and can be placed at any 3D position and oriented by a pan-tilt head. The method is based on a mixed exhaustive search and hill climbing optimisation, and outputs the next position in usable time. Results are shown with a simulated mobile range sensor navigating in CAD models of environments (closed rooms).

This paper shows that adding 3D depth information to RGB colour images improves segmentation of p... more This paper shows that adding 3D depth information to RGB colour images improves segmentation of pigmented and non-pigmented skin lesion. A regionbased active contour segmentation approach using a statistical model based on the level-set framework is presented. We consider what kinds of properties (e.g., colour, depth, texture) are most discriminative. The experiments show that our proposed method integrating chromatic and geometric information produces segmentation results for pigmented lesions close to dermatologists and more consistent and accurate results for non-pigmented lesions.