Shereen Fouad

Incorporating Privileged Information Through Metric Learning

In some pattern analysis problems, there exists expert knowledge, in addition to the original data involved in the classification process. The vast majority of existing approaches simply ignore such auxiliary (privileged) knowledge. Recently a new paradigm-learning using privileged information-was introduced in the framework of SVM+. This approach is formulated for binary classification and, as typical for many kernel-based methods, can scale unfavorably with the number of training examples. While speeding up training methods and extensions of SVM+ to multiclass problems are possible, in this paper we present a more direct novel methodology for incorporating valuable privileged knowledge in the model construction phase, primarily formulated in the framework of generalized matrix learning vector quantization. This is done by changing the global metric in the input space, based on distance relations revealed by the privileged information. Hence, unlike in SVM+, any convenient classifier can be used after such metric modification, bringing more flexibility to the problem of incorporating privileged information during the training. Experiments demonstrate that the manipulation of an input space metric based on privileged data improves classification accuracy. Moreover, our methods can achieve competitive performance against the SVM+ formulations.

Adaptive metric learning vector quantization for ordinal classification

Many pattern analysis problems require classification of examples into naturally ordered classes. In such cases, nominal classification schemes will ignore the class order relationships, which can have a detrimental effect on classification accuracy. This article introduces two novel ordinal learning vector quantization (LVQ) schemes, with metric learning, specifically designed for classifying data items into ordered classes. In ordinal LVQ, unlike in nominal LVQ, the class order information is used during training in selecting the class prototypes to be adapted, as well as in determining the exact manner in which the prototypes get updated. Prototype-based models in general are more amenable to interpretations and can often be constructed at a smaller computational cost than alternative nonlinear classification models. Experiments demonstrate that the proposed ordinal LVQ formulations compare favorably with their nominal counterparts. Moreover, our methods achieve competitive performance against existing benchmark ordinal regression models.

Learning using privileged information in prototype based models

In some pattern analysis problems, there exists expert knowledge, in addition to the original data involved in the classification process. Most of existing approaches ignore such auxiliary (privileged) knowledge. Recently a new learning paradigm - Learning Using Hidden Information - was introduced in the SVM+ framework. This approach is formulated for binary classification and, as typical for many kernel based methods, can scale unfavorably with the number of training examples. In this contribution we present a more direct novel methodology, based on a prototype metric learning model, for incorporation of valuable privileged knowledge. This is done by changing the global metric in the input space, based on distance relations revealed by the privileged information. Our method achieves competitive performance against the SVM+ formulations. We also present a successful application of our method to a large scale multi-class real world problem of galaxy morphology classification.

Ordinal-based metric learning for learning using privileged information

Learning Using privileged Information (LUPI), originally proposed in [1], is an advanced learning paradigm that aims to improve the supervised learning in the presence of additional (privileged) information, available during training, but not in the test phase. We present a novel metric learning methodology that is specially designed for incorporating privileged information in ordinal classification tasks, where there is a natural order on the set of classes. This is done by changing the global metric in the input space, based on distance relations revealed by the privileged information. The proposed model is formulated in the context of ordinal prototype based classification with metric adaptation. Unlike the existing nominal version of LUPI in prototype models [8], [9], in ordinal classifications the proposed LUPI model takes explicitly into account the class order information during the input space metric learning. Experiments demonstrate that incorporating privileged information via the proposed ordinal-based metric learning can improve the ordinal classification performance.

Automatic thresholding from the gradients of region boundaries

We present an approach for automatic threshold segmentation of greyscale images. The procedure is inspired by a reinterpretation of the strategy observed in human operators when adjusting thresholds manually and interactively by means of ‘slider’ controls. The approach translates into two methods. The first one is suitable for single or multiple global thresholds to be applied globally to images and consists of searching for a threshold value that generates a phase whose boundary coincides with the largest gradients in the original image. The second method is a variation, implemented to operate on the discrete connected components of the thresholded phase (i.e. the binary regions) independently. Consequently, this becomes an adaptive local threshold procedure, which operates relative to regions, rather than to local image subsets as is the case in most local thresholding methods previously published. Adding constraints for specifying certain classes of expected objects in the images can improve the output of the method over the traditional ‘segmenting first, then classify’ approach.

Ontological Levels in Histological Imaging

This research is supported by EPSRC through funding under grant EP/M023869/1 “Novel context-based segmentation algorithms for intelligent microscopy”.

Mereotopological Correction of Segmentation Errors in Histological Imaging

In this paper we describe mereotopological methods to programmatically correct image segmentation errors, in particular those that fail to fulfil expected spatial relations in digitised histological scenes. The proposed approach exploits a spatial logic called discrete mereotopology to integrate a number of qualitative spatial reasoning and constraint satisfaction methods into imaging procedures. Eight mereotopological relations defined on binary region pairs are represented as nodes in a set of 20 directed graphs, where the node-to-node graph edges encode the possible transitions between the spatial relations after set-theoretic and discrete topological operations on the regions are applied. The graphs allow one to identify sequences of operations that applied to regions of a given relation, and enables one to resegment an image that fails to conform to a valid histological model into one that does. Examples of the methods are presented using images of H&E-stained human carcinoma cell line cultures.

Prototype based modelling for ordinal classification

Many pattern analysis problems require classification of examples into naturally ordered classes. In such cases nominal classification schemes will ignore the class order relationships, which can have detrimental effect on classification accuracy. This paper introduces a novel ordinal Learning Vector Quantization (LVQ) scheme, with metric learning, specifically designed for classifying data items into ordered classes. Unlike in nominal LVQ, in ordinal LVQ the class order information is utilized during training in selection of the class prototypes to be adapted, as well as in determining the exact manner in which the prototypes get updated. Prototype based models are in general more amenable to interpretations and can often be constructed at a smaller computational cost than alternative non-linear classification models. Experiments demonstrate that the proposed ordinal LVQ formulation compares favorably with its nominal counterpart. Moreover, our method achieves competitive performance against existing benchmark ordinal regression models.

Epithelium and Stroma Identification in Histopathological Images Using Unsupervised and Semi-Supervised Superpixel-Based Segmentation

We present superpixel-based segmentation frameworks for unsupervised and semi-supervised epithelium-stroma identification in histopathological images or oropharyngeal tissue micro arrays. A superpixel segmentation algorithm is initially used to split-up the image into binary regions (superpixels) and their colour features are extracted and fed into several base clustering algorithms with various parameter initializations. Two Consensus Clustering (CC) formulations are then used: the Evidence Accumulation Clustering (EAC) and the voting-based consensus function. These combine the base clustering outcomes to obtain a more robust detection of tissue compartments than the base clustering methods on their own. For the voting-based function, a technique is introduced to generate consistent labellings across the base clustering results. The obtained CC result is then utilized to build a self-training Semi-Supervised Classification (SSC) model. Unlike supervised segmentations, which rely on large number of labelled training images, our SSC approach performs a quality segmentation while relying on few labelled samples. Experiments conducted on forty-five hand-annotated images of oropharyngeal cancer tissue microarrays show that (a) the CC algorithm generates more accurate and stable results than individual clustering algorithms; (b) the clustering performance of the voting-based function outperforms the existing EAC; and (c) the proposed SSC algorithm outperforms the supervised methods, which is trained with only a few labelled instances.

Unsupervised Superpixel-based Segmentation of Histopathological Images with Consensus Clustering

We present a framework for adapting consensus clustering methods with superpixels to segment oropharyngeal cancer images into tissue types (epithelium, stroma and background). The simple linear iterative clustering algorithm is initially used to split-up the image into binary superpixels which are then used as clustering elements. Colour features of the superpixels are extracted and fed into several base clustering approaches with various parameter initializations. Two consensus clustering formulations are then used, the Evidence Accumulation Clustering (EAC) and the voting-based function. They both combine the base clustering outcomes to obtain a single more robust consensus result. Unlike most unsupervised tissue image segmentation approaches that depend on individual clustering methods, the proposed approach allows for a robust detection of tissue compartments. For the voting-based consensus function, we introduce a technique based on image processing to generate a consistent labelling scheme among the base clustering outcomes. Experiments conducted on forty five hand-annotated images of oropharyngeal cancer tissue microarray cores show that the ensemble algorithm generates more accurate and stable results than individual clustering algorithms. The clustering performance of the voting-based consensus function using our re-labelling technique also outperforms the existing EAC.