Matthias S. Keil

Separating the chaff from the wheat: possible origins of the oblique effect

The oblique effect refers to a better perception of horizontal and vertical image features as compared with the perception at oblique angles. This effect can be observed in both animals and humans. Recent neurophysiological data suggest that the basis of this effect lies in the structure of the primary visual cortex, where more cortical area is devoted to processing contours with angles at horizontal and vertical orientations (cardinal orientations). It has been suggested that this cortical feature has developed according to the statistical properties of natural scenes. To examine this hypothesis in more detail, we established six image classes and categorized the images with respect to their semantical contents. From the images the oriented energy was calculated by using the corresponding power spectra. We defined simple measures for the degree (cardinal versus oblique energy ratio) and the skewness or anisotropy (aligned energy ratio) of the alignment of energy at horizontal and vertical orientations. Our results provide evidence that (1) alignment depends strongly on the environment, (2) the degree of alignment drops off characteristically at higher frequencies, and (3) in natural images there is on the average an anisotropy in the distribution of energy at the cardinal orientations (i.e., a difference between the amounts of vertical energy and horizontal energy). In light of our results, we further discuss whether the observed cortical anisotropy has its origin in phylogeny or ontogeny.

A bioinspired collision detection algorithm for VLSI implementation

In this paper a bioinspired algorithm for collision detection is proposed, based on previous models of the locust (Locusta migratoria) visual system reported by F.C. Rind and her group, in the University of Newcastle-upon-Tyne. The algorithm is suitable for VLSI implementation in standard CMOS technologies as a system-on-chip for automotive applications. The working principle of the algorithm is to process a video stream that represents the current scenario, and to fire an alarm whenever an object approaches on a collision course. Moreover, it establishes a scale of warning states, from no danger to collision alarm, depending on the activity detected in the current scenario. In the worst case, the minimum time before collision at which the model fires the collision alarm is 40 msec (1 frame before, at 25 frames per second). Since the average time to successfully fire an airbag system is 2 msec, even in the worst case, this algorithm would be very helpful to more efficiently arm the airbag system, or even take some kind of collision avoidance countermeasures. Furthermore, two additional modules have been included: a Topological Feature Estimator and an Attention Focusing Algorithm. The former takes into account the shape of the approaching object to decide whether it is a person, a road line or a car. This helps to take more adequate countermeasures and to filter false alarms. The latter centres the processing power into the most active zones of the input frame, thus saving memory and processing time resources.

A 3D-elastography-guided system for laparoscopic partial nephrectomies

We present an image-guided intervention system based on tracked 3D elasticity imaging (EI) to provide a novel interventional modality for registration with pre-operative CT. The system can be integrated in both laparoscopic and robotic partial nephrectomies scenarios, where this new use of EI makes exact intra-operative execution of pre-operative planning possible. Quick acquisition and registration of 3D-B-Mode and 3D-EI volume data allows intra-operative registration with CT and thus with pre-defined target and critical regions (e.g. tumors and vasculature). Their real-time location information is then overlaid onto a tracked endoscopic video stream to help the surgeon avoid vessel damage and still completely resect tumors including safety boundaries. The presented system promises to increase the success rate for partial nephrectomies and potentially for a wide range of other laparoscopic and robotic soft tissue interventions. This is enabled by the three components of robust real-time elastography, fast 3D-EI/CT registration, and intra-operative tracking. With high quality, robust strain imaging (through a combination of parallelized 2D-EI, optimal frame pair selection, and optimized palpation motions), kidney tumors that were previously unregistrable or sometimes even considered isoechoic with conventional B-mode ultrasound can now be imaged reliably in interventional settings. Furthermore, this allows the transformation of planning CT data of kidney ROIs to the intra-operative setting with a markerless mutual-information-based registration, using EM sensors for intraoperative motion tracking. Overall, we present a complete procedure and its development, including new phantom models - both ex vivo and synthetic - to validate image-guided technology and training, tracked elasticity imaging, real-time EI frame selection, registration of CT with EI, and finally a real-time, distributed software architecture. Together, the system allows the surgeon to concentrate on intervention completion with less time pressure.

Gradient representation and perception in the early visual system—A novel account of Mach band formation

Recent evidence suggests that object surfaces and their properties are represented at early stages in the visual system of primates. Most likely invariant surface properties are extracted to endow primates with robust object recognition capabilities. In real-world scenes, luminance gradients are often superimposed on surfaces. We argue that gradients should also be represented in the visual system, since they encode highly variable information, such as shading, focal blur, and penumbral blur. We present a neuronal architecture which was designed and optimized for segregating and representing luminance gradients in real-world images. Our architecture in addition provides a novel theory for Mach bands, whereby corresponding psychophysical data are predicted consistently.

Smooth Gradient Representations as a Unifying Account of Chevreul's Illusion, Mach Bands, and a Variant of the Ehrenstein Disk

Recent evidence suggests that the primate visual system generates representations for object surfaces (where we consider representations for the surface attribute brightness). Object recognition can be expected to perform robustly if those representations are invariant despite environmental changes (e.g., in illumination). In real-world scenes, it happens, however, that surfaces are often overlaid by luminance gradients, which we define as smooth variations in intensity. Luminance gradients encode highly variable information, which may represent surface properties (curvature), nonsurface properties (e.g., specular highlights, cast shadows, illumination inhomogeneities), or information about depth relationships (cast shadows, blur). We argue, on grounds of the unpredictable nature of luminance gradients, that the visual system should establish corresponding representations, in addition to surface representations. We accordingly present a neuronal architecture, the so-called gradient system, which clarifies how spatially accurate gradient representations can be obtained by relying on only high-resolution retinal responses. Although the gradient system was designed and optimized for segregating, and generating, representations of luminance gradients with real-world luminance images, it is capable of quantitatively predicting psychophysical data on both Mach bands and Chevreuls illusion. It furthermore accounts qualitatively for a modified Ehrenstein disk.

Intraoperative registration for liver tumor ablation

Computer aided navigation augments intraoperatively gathered U/S with planning information that the doctor carries out before the intervention on a CT volume. A crucial step for the navigation is the registration between CT and U/S. Our approach consists on a landmark based registration. The correspondences between both modalities are found automatically using a graph to graph matching algorithm. Therefore, liver and vessels are previously segmented. The whole process has being tested on 15 pairs of real clinical data. The results are promising.

A hybrid surface/image-based approach to facilitate ultrasound/CT registration

Registration of intra-operative ultrasound with preoperative CT is highly desirable as a navigational aid for surgeons and interventional radiologists. Image-based solutions generally achieve poor results due to substantially different image appearance of ultrasound and CT. A method is presented that uses surface information and tracked ultrasound to improve registration results. Tracked ultrasound is combined with surface and image-based registration techniques to register ultrasound to CT. Surface data is acquired using an optically tracked range sensor, for example time-of-flight camera. Range data is registered to CT using robust point-set registration; this registration provides an approximate transformation from tracker to CT coordinates. The ultrasound probe is also optically tracked. The probe position and surface-based registration provide a first estimate for the position of the ultrasound image in CT coordinates. This estimate is subsequently refined by a final image-based registration stage. Initial tests using Coherent Point Drift algorithm for registering surface data to CT show favorable results. Tests using both simulated and real time-of-flight range data have good convergence over a wide initial translation and rotation misalignment domain. Preliminary testing using time-of-flight surface data suggests that surface to CT registration may be useful as an initial guess enabling later more precise (but less robust) image based methods for registering ultrasound images to CT. We believe this method will enable image-based algorithms to robustly converge to an optimal registration solution.

Neural mechanisms for segregation and recovering of intrinsic image features

We present a single-scale architecture for both segregation and recovering of intrinsic image features and brightness perception. Specifically, a given intensity (or grey scale) image is first analyzed for texture (here defined as small-scale even symmetric features), surfaces (small-scale odd symmetric features) and gradients (large-scale even and odd symmetric features). In this way the image is segregated. Subsequently, textures, surfaces and gradients are recovered by corresponding neural circuits. The proposed architecture may serve as a generic building block for a variety of early vision tasks such as, for example, denoising, efficient coding, as well as mid-level tasks that build on the results from the preceding processing stages.

A neurodynamical retinal network based on reaction-diffusion systems

A dynamical model for retinal processing is presented. The model describes the output of retinal ganglion cells whose receptive field is composed of a center and a surround combining linearly. However, in comparison to the classical difference-of-Gaussian (DOG) model, center and surround are generated in two separate layers of reaction-diffusion systems, through a difference in the speed of activity-propagation between both layers. Thus, intra-layer coupling is based exclusively on next-neighbor interactions. This makes the model suitable for VLSI implementation. Furthermore, the layers are connected by equations with feedback-inhibition to form ON-center/OFF-surround and OFF-center/OFF-surround receptive fields. The models output in the early dynamics corresponds to high-resolution contrast information, whereas the output at later times can be considered as correlated with local brightness and darkness, respectively. To examine this in more detail, simulations with the Hermann/Hering-grid and grating induction were carried out.

How is luminance information passed into the cortex: emergent multifunctional behavior of a simple cell model

In modeling brightness perception, one problem of high biological relevance is how luminance information is transmitted into the primary visual cortex. This is especially interesting in the light of recent neurophysiological studies, which suggest that simple cells are responding shallowly to homogeneous illuminated surfaces. This indicates that simple cells possess far more functional complexity as the wide-spread notion of mere line and edge detectors. Here we present new neural circuits for modeling even and odd simple cells, capable of transmitting brightness information without using an extra `luminance- channel. Although these circuits taken for themselves can not be regarded yet as a full brightness model, however, they might gain some insight in why the visual system is using certain processing strategies. These include e.g. the segregation in ON and OFF channels and the mutual inhibition of simple cell pairs which are in anti-phase relation. These simple cell circuits turn out to be robust against noise, and thus might find its application in a border detection scheme, beside of being a building block for a more sophisticated brightness-model.