ITU/Image: Group Meetings

On Wednesdays at 12 o'clock we have a lunch meeting. The meeting is mostly on general, administrative and organizational issues concerning the group. As part of the meeting, research talks will be given on occasions.

At 11 o'clock preceding the lunch meetings are some purely scientific meetings with discussion. Every other week they are on medical image analysis and the weeks in between they are on foundations of image analysis.

The scientific meetings are held in 4A09 if not otherwise indicated, and the lunch meetings take place in room 4D01 (middle area of section 4D).

Passed talks

2005-08-09		Mads Nielsen: Image Group Strategy Mads has written his conclusions on the strategy day June 3rd. These may be altered by subsequent discussions. This lunch meeting will be devoted to a discussion of the draft. [Get the Draft]
2005-04-12		Discussion on MMT courses and their interdependency The discussion will include: Quick update on new MMT line. Major revision of MMsys. Should we suggest a merger of SP and Image for one first semester course. In relation to 2 and in general how about more advanced stuff such as CV, foundations and goodies from current IA. One course, two alternating? Graphics. What do we think about the rest of the current MMT.
2005-04-05 Room 4A05		Thomas Hansen: Signal processing at DK-Technologies DK-Technologies develop, manufacture and distribute audio and video equipment for the professional and industrial markets, e.g. equipment used in radio and television broadcast, audio and video production houses, the movie industry, etc. The presentation will have the following agenda: General presentation of DK-Technologies. (10min) Demonstration of equipment. (10min) Discussion of some of the issues related to digital video processing. (10min)
2005-02-22		Frederik Brinck, DTU: (Master's thesis on signal processing)
2005-02-15		Lewis Griffin (University College London): Geometric Texton Theory: The 1-D, 2nd-order jet Much is understood of how quantitative aspects of image structure are measured by V1 simple cells, but less about how qualitative structure is determined from these measurements. Geometric texton theory (GTT) aims to describe this step from quantitative to qualitative. GTT (Griffin et al., 2004, Vision Research 44, pp. 407–421) proposes that qualitative feature categories arise through consideration of the maximum likelihood (ML) explanations of image measurements. This posits that a pair of output vectors of an ensemble of co-localised neurons signal the same feature category if and only if the corresponding ML explanations are qualitatively similar. I have studied the mathematics and image statistics relevant to GTT for the limited case of measurement by 1-D Gaussian derivative filters of up to second order. Mathematical analysis has identified the simplest explanations for measurements by such filters, while empirical results have identified the ML. The ML explanations are not the most simple under any of the definitions of “simple” that have been examined. However, the ML explanations are simple and do have properties predicted by GTT. In particular, they change rapidly and qualitatively for certain narrow regions of measurement space, while remaining qualitatively extremely stable between those transition regions. Three feature categories arise naturally from the data: light bars, dark bars, and edges. The results are consistent with GTT.
2005-01-11		Arjan Kuijper: Symmetry sets and pre-symmetry sets teaser.
2004-11-30		Mads Nielsen: Challenges in Computer Vision (continued).
2004-11-23		Martin Wallengren Nilsson and Andreas Rishede Hyllested: Semi-automatic foreground extraction Our thesis is done in cooperation with Laerdal-Sophus A/S (LS). The core products of LS are simulators for training of medical personal. To create realistic simulators LS uses a large number of natural images that has to be segmented. This is at the moment done manually. Our task is to create a semiautomatic segmentation tool to make the segmentation task more efficient. At the moment we are working on a method based on energy minimization via graph cut. In the talk we will give a short introduction to graph cut and then give a more detailed description of our implementation. One of our challenges is that LS have very large images (3072x2048). To reduce the computation time we make a pre-segmentation using a multi-scale based watershed algorithm. The nodes in the graph are then built on small segments and not on pixels. Our implementation is done in C++.
2004-11-09		Arjan Kuijper: Geometric Skeletonization using the Symmetry Set In this paper we present a novel method to derive the skeleton of a shape by means of its Symmetry Set. We use the property that the Medial Axis is a subset of the Symmetry Set. This set is derived by means of zero crossings of two equations, based on geometric arguments. The resulting skeleton can easily be annotated with respect to main axis and branches, since this information follows directly from the Symmetry Set.
2004-11-02		Mads Nielsen: Challenges in Computer Vision.
2004-10-26		Mac Wendelboe: Using the “Mosix” cluster and the queuing system I will give a short explaination of how you log into the cluster, transfer files, issue computing jobs, make an interactive connection to a Windows node and how you monitor job and cluster status. Have a look at the cluster guide and the queue system guide. Slides are available.
2004-10-19		Michael Lund: The STAPLE Algorithm.
2004-10-05		Uffe Damgaard Pedersen: Modelling Human Embryos Using a Variational Level Set Approach At fertility clinics, living human embryos are evaluated by visual inspection of light microscopy images. The criteria to assess the quality are number of blastomeres (cells) of the embryo, regularity of size and degree of fragmentation. Today, this evaluation process is based on interpretatio of 3D information projected to 2D images. It is believed that an overall improvement in the process could be achived, if the process could be based on 3D information with quantitative measurements instead. Unfortunately, due to limitations in the acquisition process, the 3D information cannon be obtained directly. Previous work on building 3D models has used solid spheres to simulate the position and size of the blastomeres. This thesis' main objective is to make use of the level set methodology to address this problem. A new 3D model is developed by using a variational level set approach to model the blastomeres and the interior of the Zona Pellucida as elastic objects with individual surface tensions, thereby enabling them to deform due to mutual object interaction. Positional information from microscopy images is embedded as soft constraints requiring that an obtained curve must be located on an object's surface and that the object must be located inside a bounding generalized cylinder passing through this curve. There is no quantitative standard to measure the results against, but results are promising and correspond well to human interpretation of the light microscopy images. Find demonstrations and the full text thesis on the Master's Theses homepage.
2004-09-21		Nanna Glerup: Spherical harmonics If objects are known as points on the surface of the objects, it is difficult to compare the objects. The number of points for two objects might not be the same and the corresponce between the points might not be known. If the objects are represented by a large number of points, the dimensions of the representation space is reduced, if the objects are projected into the spherical harmonics space. The talk is about the spherical harmonic functions, how to project an object and the calculatation made on the objects in the spherical harmonic space.
2004-09-07		Jenny Folkesson: Locating Articular Cartilage in MR Images Accurate assessment of the articular cartilage is of great importance when diagnosing and monitoring the progress of joint diseases such as osteo-arthritis. A fully automated cartilage assessment method is preferable compared to methods using manual interaction in order to avoid inter- and intra-observer variability. A first step in this assessment is to locate the cartilage in MRI of the knee. This is achieved using a kNN-classifier, were voxels are classified as cartilage or non-cartilage based on prior knowledge of the cartilage structure. Currently, the classification scheme finds a mean cartilage location that is a few millimeters away from the real cartilage center of mass. The found cartilage location will be used for initiating a deformable shape model that will be fitted to the cartilage, and then the shape model will be used in the cartilage evaluation. This work is a collaboration between ITU and CCBR, and will involve approximately 350 MR scans.
2004-07-20		Ann Lee: Manifold learning by geometric diffusion
2004-06-29		Rein van den Boomgard: Robust Estimation of Local Image Structure In characterizing a local image patch with its Gaussian derivatives up to order N (the N jet) we are effectively constructing the Taylor expansion of the _blurred_ image. The Gaussian facet model constructs a polynomial model of the unblurred image by fitting a polynomial to the image data. It turns out that the coefficients in the facet model are linear combination of the Gaussian derivatives. The Gaussian facet model is based on a least squares fitting procedure. In case a local neighborhood shows two different structures these will get mixed up in the estimation procedure and the resulting model will be blurred. A robust estimation procedure replaces the quadratic error norm with a norm that puts less emphasis on outliers and thus leads to better estimates of the model parameters (i.e. less blurring). We will show that the robust estimation of a local polynomial model is closely related to the local mode finding algorithm (and thus the mean shift algorithm) and to the bilateral filter. We will also describe robust estimation procedures to estimate local orientation in images and to estimate the optic flow. Both these procedures are based on a 'robustification' of the structure tensor.
2004-06-15 Room 2.03		Mads Nielsen talked about short time memory and the possibilities of collaboration between psycology and computer vision.
2004-05-25 Room 2.03		Lars Arne-Conrad Hansen: Quantizing Calcification in the Lumbar Aorta on 2-D Lateral X-Ray Images We seek to improve upon the standard method of assessing the degree of calcification in the lumbar aorta, which is commonly used on lateral 2-D x-rays. The necessity for improvement arises from the fact that the existing method can not measure subtle progressions in the plaque development; neither is it possible to express the density of individual plaques. Both of these qualities would be desireable to assess, since they are the key for making progression studies as well as for testing the effect of drugs in longitudinal studies. Our approach is based on inpainting, a technique used in image restoration as well as postprocessing of film. In this study we discuss the potential implications of total variation inpainting for characterizing aortic calcification.
2004-04-27 Room 2.03		Erik Bjørnager Dam: Shape Model Regularization The use of statistical shape models in medical image analysis is growing due to the ability to incorporate prior organ shape knowledge for tasks such as segmentation, registration, and classification. Shape models are trained from collections of segmented organs. Though manual interaction during training can ensure correspondence, it also introduces bias and ruins reproducibility --- automation is desirable. We present a novel shape model construction method via a medial shape representation. The automatic method is based on an iterative bootstrap method that alternates between shape representation optimization and analysis of shape mean and variations. The method is used to create a model from 46 segmented prostates with quantitatively and intutitively good results.
2004-04-20 Room 2.03		Kim Stenstrup Pedersen and Martin Lillholm: Jet Based Feature Classification In this paper, we investigate to which extent the "raw" mapping of Taylor series coefficients into jet-space can be used as a "language" for describing local image structure in terms of geometrical image features. Based on empirical data from the van Hateren database, we discuss modelling of probability densities for different feature types, calculate feature posterior maps, and finally perform classification or simultaneous feature detection in a Bayesian framework. We introduce the Brownian image model as a generic background class and extend with empirically estimated densities for edges and blobs. We give examples of simultaneous feature detection across scale.
2004-04-13 Room 2.03		Maz Spork will present a possible infrastructure for analysis and partial replacement of streaming video sourced from live broadcasts. The aim is to uncover the challenges of real-time replacements of, e.g., advertisements or artifacts embedded in image sequences given different sets of meta-information about the streamed data. Maz Spork holds an M.Sc. in computer science from the University of Copenhagen and works with product development of distributed systems within the digital tv industry.
2004-03-30 Room 2.55		Ole Fogh Olsen: Generic properties for continous watersheds In this work we analysis and relate the generic structure of continouos watershed to other well-established differential feature detectors such edge detectors and corner detectors. We prove that the classical edges similar to Canny edges will generically intersect the continous watersheds and characterize the differential structure of these points. We also prove the generic relations between watershed junctions and corner detector based on curvature. This analysis is feasiable by exploiting the fact that the continouos watersheds are a subset of the separatrix in a continuous image.
2004-03-23 Room 2.55		Arjan Kuijper: Presymmetry sets In shape analysis the Symmetry Set (SS) and a subset of it, the Medial Axis, can be used to represent the shape. The Symmetry Set is defined as the closure of circles tangent to the shape at at least two points. As the result may live on an unbounded domain in case of concave shapes, one can also investigate the pre-SS. This set is defined as the points at which the circle is tangent. This set exhibits nice properties which I will reveal during the presentation.
2004-03-09 Room 2.55		Ole Fogh Olsen: Structure of optic flow The optic flow field is defined such that along integral lines of the field the image intensity remains constant. For each time instance in an image sequence poles are created in the optic flow field at the position of spatial image singularities. We describe the generic flow singularities and the generic transitions of these over time. For classic analytic flow fields the classification of the generic topology is based on points of vanishing flow which can be further subdivided into repellers, attractors, whirls, and combinations hereof. We point out the resemblance, but also the important differences between the structure of the classical analytic flow field, and the structure of the optic flow field expressed through its normal flow. We conclude by giving an operational scheme for the detection of these singularities and events; and apply the scheme to two different examples within respectively attention mechanism and the degree of turbulence in a flow field.
2004-03-02 Room 2.55		Marleen de Bruijne: Shapes
2004-02-10 Room 2.03		Nanna Glerup: Asymmetry Quantization and Application to Human Mandibles All biological objects exhibit some degree of asymmetry, but for some parts of the human body, excessive asymmetry is a sign of pathology. Hence, the problem is to draw the line between categorization of objects being too asymmetric and objects exhibiting normal asymmetry. With a measure of asymmetry, the statistics on asymmetry for normal and pathological anatomical structures can be compared. Symmetry is a well-known mathematical group theoretical concept. In this paper, we will mathematically define the concept of weak symmetry, including topological symmetry, which serves as a basis for quantizing asymmetry. The methodology is based on non-rigid registration in the sense that the "size" of a diffeomorphism describes the amount of asymmetry. We will define this size in terms of the minimum biological work needed. That is, we evaluate how much work the biological system must carry out in order to make the object symmetrical; or identically, how much work has been carried out in order to make the ideal symmetrical object into the current (slightly) asymmetrical object. The quantization of asymmetry is validated on a set of normal (assumed near symmetrical) mandibles, and a set of pathological assumed non-symmetric mandibles exhibiting a statistically significant increase of asymmetry.
2004-02-03 Room 2.03		Jakob Raundahl: Effect of projective viewpoint in detecting temporal density changes An important question in mammographic image analysis is the importance of the projected view of the breast. Can temporal changes in density be detected equally well using either one of the commonly available views Medio-Lateral (ML) and Cranio-Caudal (CC) or a combination of the two? Two sets of mammograms of 50 patients in a double-blind, placebo controlled hormone replacement therapy (HRT) experiment were used. One set of ML and CC view from 1999 and one from 2001. HRT increases density which means that the degree of separation of the populations (one group receiving HRT and the other placebo) can be used as a measure of how much density change information is carried in a particular view or combination of views. Earlier results have shown a high correlation between CC and ML views leading to the conclusion that only one of them is needed for density assessment purposes. A similar high correlation coefficient was observed in this study (0.85), while the correlation between changes was a bit lower (0.71). Using both views to separate the patients receiving hormones from the ones receiving placebo increased the area under corresponding ROC curves from 0.76 ± 0.04 to 0.79 ± 0.04.
2004-01-27 Room 2.03		Lars Arne-Conrad Hansen: Prediction of the Location of the Lumbar Aorta using the First Four Lumbar Vertebrae as a Predictor This paper is one of the first steps towards the development of a mass-screening tool, well-suited for quantizing the extend of calcific deposits in the lumbar aorta, which should deliver reliable and easily reproducible data. The major problem is that non-calcified parts of the aorta are not visible on conventional x-ray images. We investigate whether or not it is possible to predict the location of the lumbar aorta, using the first four lumbar vertebrae as prior. We build a conditional probabilistic model from 90 manually annotated datasets. Using this model we made inferences on the position of the aortic walls given the position and shape of the four vertebrae. Of particular interest is the performance of the probabilistic model in comparison to the mean aortic shape. Due to the fact that our data set for this particular study only contained 90 hand-annotated images, we evaluated the model using the "leave-one-out" method. The resulting distance from the predicted to the actual aorta was then compared to the distance from the mean aorta to the actual aorta. The obtained results are encouraging; our conditional model provides results that are up to 38 % better than the prediction using only the mean shape, and yields an overlap index of 0.89, whereas the mean shape only produces 0.83.
2004-01-20 Room 2.03		Erik Bjørnager Dam and Marco Loog: Integrating Automatic and Interactive Brain Tumor Segmentation We integrate automatic segmentation based on supervised learning with an interactive multi-scale watershed segmentation method. The combined method automatically provides an initial segmentation that applies the building blocks that the user can use in the interactive method. Thereby the two approaches are seamlessly integrated and the combined method can be used on the full range of problems from very easy to very difficult segmentation tasks resulting in different levels of interaction needed. The method is evaluated for segmentation of brain tumors.