Stephan Al-Zubi

Using an Active Shape structural model for biometric sketch recognition

A deformable shape model called Active Shape Structural Model (ASSM) is used within a biometric framework to define a biometric sketch recognition algorithm. Experimental results show that mainly structural relations rather than statistical features can be used to recognize sketches of different users with high accuracy.

Segmentation of multiple sclerosis lesions from MR brain images using the principles of fuzzy-connectedness and artificial neuron networks

Segmentation is an important step for the diagnosis of multiple sclerosis. In this paper, a method for segmentation of multiple sclerosis lesions from magnetic resonance (MR) brain image is proposed. The proposed method combines the strengths of two existing techniques: fuzzy connectedness and artificial neural networks. From the input MR brain image, the fuzzy connectedness algorithm is used to extract segments which are parts of cerebrospinal fluid (CSF), white matter (WM) or gray matter (GM). Segments of the MRI image which are not extracted as part of CSF, WM or GM are processed morphologically, and features are computed for each of them. Then these computed features are fed to a trained artificial neural network, which decides whether a segment is a part of a lesion or not. The results of our method show 90% correlation with the experts manual work.

Deformable structural models

A hierarchical framework for the recognition of complex deformable shapes is developed. In extension to traditional approaches an additional layer of control is introduced to guide the local search for subshapes. This is realized by incorporating knowledge about their spatial relationships. A new technique of expectation maps is applied to allow simultaneous shape searches to influence each other. Furthermore, these maps are used to assess spatial coherence among shapes. Thus, the occurrence of well matched shapes at some places in the image may suggest searches for related shapes at other positions. An application to classify species in ant image databases shows promising initial results.

Extending Active Shape Models to Incorporate a-priori Knowledge about Structural Variability

A new deformable shape model is defined with the following properties: (1) A-priori knowledge describes shapes not only by statistical variation of a fixed structure like active shape/appearance model but also by variability of structure using a production system. (2) Multiresolution description of shape structures enable more constrained statistical variation of shape as the model evolves in fitting the data. (3) It enables comparison between different shapes as well as characterizing and reconstructing instances of the same shape. Experiments on simulated 2D shapes demonstrate the ability of the algorithm to find structures of different shapes and also to characterize the statistical variability between instances of the same shape.

Fusing Markov random fields with anatomical knowledge and shape-based analysis to segment multiple sclerosis white matter lesions in magnetic resonance images of the brain

This paper proposes an image analysis system to segment multiple sclerosis lesions of magnetic resonance (MR) brain volumes consisting of 3 mm thick slices using three channels (images showing T1-, T2- and PD -weighted contrast). The method uses the statistical model of Markov Random Fields (MRF) both at low and high levels. The neighborhood system used in this MRF is defined in three types: (1) Voxel to voxel: a low-level heterogeneous neighborhood system is used to restore noisy images. (2) Voxel to segment: a fuzzy atlas, which indicates the probability distribution of each tissue type in the brain, is registered elastically with the MRF. It is used by the MRF as a-priori knowledge to correct miss-classified voxels. (3) Segment to segment: Remaining lesion candidates are processed by a feature based classifier that looks at unary and neighborhood information to eliminate more false positives. An experts manual segmentation was compared with the algorithm.

Generalizing the Active Shape Model by Integrating Structural Knowledge to Recognize Hand Drawn Sketches

We propose a new deformable shape model Active Shape Structural Model (ASSM) for recognition and reconstruction. The main features of ASSM are: (1) It describes variations of shape not only statistically as Active shape/Appearance model but also by structural variations. (2) Statistical and structural prior knowledge is integrated resulting in a multi-resolution shape description such that the statistical variation becomes more constrained as structural information is added. Experiments on hand drawn sketches of mechanical systems using electronic ink demonstrate the ability of the deformable model to recognize objects structurally and reconstruct them statistically.

Imitation Learning and Transferring of Human Movement and Hand Grasping to Adapt to Environment Changes

We propose a model for learning the articulated motion of human arm and hand grasping. The goal is to generate plausible trajectories of joints that mimic the human movement using deformation information. The trajectories are then mapped to a constraint space. These constraints can be the space of start and end configuration of the human body and task-specific constraints such as avoiding an obstacle, picking up and putting down objects. Such a model can be used to develop humanoid robots that move in a human-like way in reaction to diverse changes in their environment and as a priori model for motion tracking. The model proposed to accomplish this uses a combination of principal component analysis (PCA) and a special type of a topological map called the dynamic cell structure (DCS) network. Experiments on arm and hand movements show that this model is able to successfully generalize movement using a few training samples for free movement, obstacle avoidance and grasping objects. We also introduce a method to map the learned human movement to a robot with different geometry using reinforcement learning and show some results.

Learning to mimic motion of human arm and hand grabbing for constraint adaptation

We propose a model for learning the articulated motion of human arm and hand grabbing. The goal is to generate plausible trajectories of joints that mimic the human movement using deformation information. The trajectories are then mapped to a constraint space. These constraints can be the space of start and end configuration of the human body and task-specific constraints such as avoiding an obstacle, picking up and putting down objects. Such a model can be used to develop humanoid robots that move in a human-like way in reaction to diverse changes in their environment and as a priori model for motion tracking. The model proposed to accomplish this uses a combination of principal component analysis (PCA) and a special type of a topological map called the dynamic cell structure (DCS) network. Experiments on arm and hand movements show that this model is able to successfully generalize movement using a few training samples for free movement, obstacle avoidance and grabbing objects.

Learning to Imitate Human Movement to Adapt to Environmental Changes

A model for learning human movement is proposed. The learning model generates plausible trajectories of limbs that mimic the human movement. The learning model is able to generalize these trajectories over extrinsic constraints. These constraints result from the space of start and end configuration of the human body and task-specific constraints such as obstacle avoidance. This generalization is a step forward from existing systems that can learn single gestures only. Such a model is needed to develop humanoid robots that move in a human-like way in reaction to diverse changes in their environment. The model proposed to accomplish this uses a combination of principal component analysis (PCA) and a special type of a topological map called the dynamic cell structure (DCS) network. Experiments on a kinematic chain of 3 joints show that this model is able to successfully generalize movement using a few training samples for both free movement and obstacle avoidance