Christoph Rasche

Polytechnic University of Bucharest | UPB · ETTI (Electronica, Telecomunicatie si Tehnologie de Informatie)

We introduce a method for region analysis that is based on analyzing the ridges (symmetric-axes) of a distance map. The detection technique for ridge pixel uses a mere local maxima search and has therefore minimal complexity. The ridges constitute a height profile, that is suitable for abstracting regions by means of a simple parameterization. The...

We introduce a contour detection method that has relatively low complexity yet still high accuracy. The method is based on extrema detection along the four principal orientations, a trick that can be used to detect not only edges, but in particular also ridges and rivers. We make a comparison to the popular Canny algorithm and show that our method'...

A database of 9405 cervical cells is introduced, which was collected from Pap-smear images: 1791 cells are pathologic cases (two types), the rest are healthy cases (three types). Their cell nuclei are classified using two methods: once with a traditional feature engineering approach using in particular iso-contours; and once with a Deep Belief Netw...

We evaluate the suitability of rapidly extracted contour and region features for a variety of tasks. Both types of features are extracted with simplest techniques and show a large degree of correspondence with bounding boxes of annotated datasets (VOC2007, CityScape), almost as large as object proposal techniques. The features are then partitioned...

An analysis of a parametric structural description is presented, with a focus on image classification. The structural description is based on contours and regions and deploys only a minimal set of 10 geometric parameters, added by chromatic and position attributes. It is evaluated on six image collections (Urban/Natural, Scene-15, Indoor-67, NR-45,...

A framework for scene categorization and identification is introduced that is based on parameterized contour and region information. Its feature extraction methodology has very low complexity: to detect edge-, ridge-and river-pixels, an extrema-detection technique is deployed that uses relational operations only without computing products; to find...

Contour and region features are rarely used in visual place recognition, although they could serve as object proposals for interaction with the environment. In this study, a system is introduced that extracts both features in less than 100 milliseconds for a 480x640 image. The features are then parameterized using geometric and appearance attribute...

A tree-growing segmentation procedure for finding regions of similar color is introduced. The procedure starts by segregating the chromatic space into two groups resulting typically in an under-segmented map. In a second step, each one of those regions is split again resulting in a second, more refined segmentation map. Through this repeated binary...

We report on a methodological framework that analyzes land-use images with engineered (manually designed) features. As older feature engineering methods suffered from excessive computation of their features, we therefore, introduce techniques that are faster and also more elaborate than any previous approach. Feature extraction and description is b...

Minimal overview of the three principal approaches to representation in computer vision (Deep Nets, Local Features, Structural Description).

A two-page summary of my approach to build a visual system. (Scroll down for text)

A list of short project descriptions.

A framework for characterizing the intensity landscape with simplest neighborhood operations is introduced. The characterization is based on contours and regions. To detect edge-, ridge-and river-pixels, an extrema-detection technique is presented that uses relational operations only. To find regions, the K-Means clustering algorithm is employed us...

This is a dense introduction to the field of computer vision. It covers all three approaches, the classical engineering approach based on contours and regions; the local-features approach; and the Deep Learning approach; summarized in Table 1. The book provides plenty of code snippets and copy-paste examples for Matlab, Python and OpenCV, the latte...

This workbook provides a rapid, practical access to the topic of pattern recognition. The emphasis lies on applying and exploring the statistical classification methods in Matlab or Python; the mathematical formulation is minimal. Plenty of code examples are given that allow to immediately play with these methods and that can serve as a reference g...

A procedure for segmentation is introduced that uses the K‐means algorithm to decompose an image by recursively applying the K‐means with k=2. This 2‐means decomposition procedure finds any region that corresponds to a scene part, but it is also prone to oversegmentation. The procedure has already been used in two tasks previously: a scene classifi...

An approach to lesion recognition is described that for lesion localization uses an ensemble of segmentation techniques and for lesion classification an exhaustive structural analysis. For localization, candidate regions are obtained from global thresholding of the chromatic maps and from applying the K-Means algorithm to the RGB image; the candida...

We introduce methods that analyze the structure within images based on an exhaustive parameterization of contour and region information. Contour analysis consists of ridge, river and edge detection carried out with a simplest extrema search; the extracted segments are partitioned and abstracted using amplitude signatures; and then grouped to form m...

This is a one-script Matlab example of how to detect ridges, rivers and edges in a gray-scale image.

A novel method for the geometric analysis and abstraction of curves is presented. The method generates a local-global (multi-resolution) amplitude space using labeling operators that identify curved, inflected and straight segments. Then, a consistency analysis along the resolution axis is carried out and several spectra are formed. The consistency...

The Babes-Papanicolaou test (also known as Pap smear) is a method of cervical cancer screening used to detect abnormal cells which are or can become cancerous. Since the visual inspection of pap smears is very time consuming, the need for automatic methods is required. This paper presents an algorithm for the automatic detection of nuclei within pa...

Satellite images are typically classified and analyzed with relatively simple features (filters). Here we introduce a methodology that uses contour information and we show that we can outperform a classification benchmark with it. The preprocessing duration is somewhat longer than mere (local) filtering , but with the obtained information we are ab...

We introduce a multiresolution analysis for detecting modes in distributions of lower dimensionality. The analysis generates a (non-parametric) density estimate for a range of bandwidths (kernel widths), thus creating a spatial-bandwidth (density) space. In this space one can detect modes by carefully analyzing the density 'consistency' along the b...

In this paper we investigate the performance of visual features in the context of video genre classification. We propose a late-fusion framework that employs color, texture, structural and salient region information. Experimental validation was carried out in the context of the MediaEval 2012 Genre Tagging Task using a large data set of more than 2...

Somewhat. We extracted, partitioned and described con tours, histogrammed their geometric parameters and concatenated the histograms to form a single image vector with which we classified the plant images using a Linear Discriminant Analysis (LDA); that is, no segmentation or saliency selection was performed. Despite the obvious simplicity of the L...

In this paper, we propose an audio-visual approach to video genre categorization. Audio information is extracted at block-level, which has the advantage of capturing local temporal information. At temporal structural level, we asses action contents with respect to human perception. Further, color perception is quantified with statistics of color di...

This paper presents an automatic algorithm for static hand gesture recognition relying on both depth and intensity information provided by a time-of-flight (ToF) camera. The combined depth and intensity information facilitates the segmentation process, even in the presence of a cluttered background (2 misses out of 450 images). Gesture classificati...

Our study explores eye movements and driving behaviour in safety-critical situations. We collected eye movements from subjects instructed to drive predetermined routes in a driving simulator. While driving, the subjects performed various cognitive tasks designed to divert their attention away from the road. The subjects were divided in two groups,...

In this paper we address the issue of automatic video genre classification and propose three categories of content descriptors. At temporal level, video content is described in terms of visual rhythm, action content and amount of gradual transitions. Further, colors are globally described using statistics of color distribution, elementary hues, col...

Spontaneous fixations onto shapes are driven by a structural analysis. But is such analysis also carried out during free viewing of real-world scenes? Here, we analyze how fixation locations in such scenes are related to their region using the region's symmetric axes as a reference. Each fixation location is compared with respect to its nearest sym...

A method for the systematic parameterization of segment alignments is presented, which was developed for the task of (general) image classification. The method starts by accurately describing the geometric alignment of pairs of segments, which in turn are used to progressively build more complex descriptors such as polygons and clusters. To minimiz...

A method for the boundary representation of 'simple' shapes is presented. It is based on the radial signature and exploits its extrema information to arrive at a low-dimensional geometric description (ca. 10 dimensions). This short description can represent shapes well, which is demonstrated on the Corel and MPEG7 collection. When this description...

We continued to test our image classificationmethodology in the photo-annotation task of the ImageCLEF competition [Nowak et al., 2011] using a visual-only ap- proach performing automated labeling - however with little algorithmic improve- ment as compared to last year. Our labeling process consisted of three phases: 1) feature extraction using col...

Words that are rated as acquired earlier in life receive shorter fixation durations than later acquired words, even when word frequency is adequately controlled (Juhasz & Rayner, 2003; 2006). Some theories posit that age-of-acquisition (AoA) affects the semantic representation of words (e. g., Steyvers & Tenenbaum, 2005), while others suggest that...

The idea of gaze guidance is to lead a viewer’s gaze through a visual display in order to facilitate the viewer’s search for specific information in a least-obtrusive manner. This study investigates saccadic orienting when a viewer is guided in a fast-paced, low-contrast letter identification task. Despite the task’s difficulty and although guiding...

Introduction: Measured human perceptual learning (or lack of) for a given task is generally attributed to properties of the human perceptual system. However, as in other domains (e.g. object recognition, Tjan and Legge, 1998), the amount of task-inherent stimulus information to be learned will also influence the amount of human perceptual learning....

We address the issue of automatic video genre retrieval. We propose three categories of content descriptors, extracted at temporal, color and structural level. At temporal level, video content is described with visual rhythm, action content and amount of gradual transitions. Colors are globally described with statistics of color distribution, eleme...

A decomposition is described, which parameterizes the geometry and appearance of contours and regions of gray-scale images with the goal of fast categorization. To express the contour geometry, a contour is transformed into a local/global space, from which parameters are derived classifying its global geometry (arc, inflexion or alternating) and de...

Numerous attempts have been made in the past to predict where observers fixate when viewing natural images (Itti et al 06, Tatler et al 05, Kienzle et al 06). Despite great efforts, it seems that the prediction rate for fixation vs. random patches is limited between a ROC-area value of 0.62–0.68. We used displays containing dynamic random noise to...

We tested our image classification methodology in the photo-annotation task of the ImageCLEF competition [Nowak, 2010] using a visual-only approach performing automated labeling. Our labeling process consisted of three phases: 1) feature extraction using color histogramming and using a novel method of structural description, that was exploited in a...

An image classification system is introduced, that is predominantly based on a description of contours and their relations. A contour is described by geometric parameters characterizing its global aspects (arc or alternating) and its local aspects (degree of curvature, edginess, symmetry). To express the relation between contours, we use a multi-di...

In many perceptual learning studies, detection and discrimination thresholds are assessed using adaptive (e.g. staircase) methods. The measured decrease in threshold is interpreted purely as an improvement in perceptual ability, assuming either an unbiased observer or temporally-invariant levels of bias. Unfortunately, this method does not allow on...

Visual orienting has typically been characterized using simple displays--for example, displays with a static target placed on a homogeneous background. In the present study, visual orienting was investigated using a dynamic broadband (1/f) noise display that should mimic a more naturalistic setting and that should allow saccadic orienting experimen...

We introduce a method to describe the geometry of contours, which is potentially benefical for biological motion perception. The method consists of systematic labeling of contour segments for different window sizes, which allows to determine high-curvature points and to derive a parametric description of contours. The method is considered as comple...

Several studies have shown that the precision of smooth pursuit eye speed can match perceptual speed discrimination thresholds during the steady-state phase of pursuit [Kowler, E., & McKee, S. (1987). Sensitivity of smooth eye movement to small differences in target velocity. Vision Research, 27, 993-1015; Gegenfurtner, K., Xing, D., Scott, B., & H...

An algorithmic framework is presented, that is suitable for image categorization and scene understanding. The framework is based on a decomposition of visual structure into elementary contour segments, which then are synthesized to form structural relations and groupings. The output of the framework can explain all (structural) pop-out phenomena as...

At slow speeds, chromatic isoluminant stimuli are perceived to move much slower than comparable luminance stimuli. We investigated whether smooth pursuit eye movements to isoluminant stimuli show an analogous slowing. Beside pursuit speed and latency, we studied speed judgments to the same stimuli during fixation and pursuit. Stimuli were either la...

An excitable membrane is described which can perform different visual tasks such as contour detection, contour propagation, image segmentation, and motion detection. The membrane is designed to fit into a neuromorphic multichip system. It consists of a single two-dimensional (2-D) layer of locally connected integrate-and-fire neurons and propagates...

Contemporary theoretical accounts of perceptual learning typically assume that observers are either unbiased or stably biased across the course of learning. However, standard methods for estimating thresholds, as they are typically used, do not allow this assumption to be tested. We present an approach that allows for this test specific to perceptu...

Prominent models of perceptual learning (e.g., Gold, Bennett, & Sekuler, 1999; Dosher & Lu, 1999) assume an internal response criterion that is stable across time. In addition, many studies use a two-interval two-alternative forced-choice (2AFC) staircase procedure that is assumed to be immune from the effects of response bias. However, this assump...

A neural architecture is presented that encodes the visual space inside and outside of a shape. The contours of a shape are propagated across an excitable neuronal map and fed through a set of orientation columns, thus creating a pattern which can be viewed as a vector field. This vector field is then burned as synaptic, directional connections int...

The symmetric-axis transform is a process that dynamically encodes the space of a visual shape through self-interaction of its contours. It is generally simulated using computer algorithms. A neural architecture of this transformation is presented that is conceptually simple enough for a hardware implementation. Its architecture consists of a wave-...

We propose a neural architecture that estimates the speed of motion. The basis is a two-dimensional map made of locally connected integrate-and-fire neurons, that propagates and integrates synaptic input in a dendritic-cable-like manner, but irrespective of any direction. The propagation dynamics of such a map are tuned to filter preferred speeds:...

The reader is presented an approach to the construction of a visual system, which is behaviorally, computationally and neurally motivated. The central goal is to characterize the process of visual categorization and to find a suitable representation format that can successfully deal with the structural variability existent within visual categories....

We describe a neuromorphic retina that signals a luminance edge as a spike. In a fast process, the luminance profile of the receptor layer determines the membrane potential of the ganglion cells and their individual, adjustable spiking thresholds. In a slower process, a wave-propagation process, the charge of ganglion cells with high membrane poten...

We try to understand the basics of human image processing from a gist recognition perspective. Because the gist is only a subset of the image's information, we think that it is extracted with help of interpretation (feedback). In a perceptual section we list possible mechanisms that the interpretation process uses to determine the gist: in addition...

An analog electronic circuit has been developed and described that adapts the electrotonic properties of a silicon dendrite. The dendrite is modeled by the method of compartmental modeling, consisting of three dendritic compartments each containing a synaptic conductance, and one somatic compartment containing a spiking mechanism. Dendritic synapti...

The recent quantitative description of activity-dependent depression in the synaptic transmission between cortical neurons has lead to many interesting suggestions of possible computational implications. Based on a simple biological model, we have constructed an analog circuit that emulates the properties of short-term depressing synapses. The circ...

We have developed an analog very-large-scale integrated (aVLSI) electronic circuit that emulates a compartmental model of a neuronal dendrite. The horizontal conductances of the compartmental model are implemented as a switched capacitor network. The transmembrane conductances are implemented as transconductance amplifiers. The electrotonic propert...

We describe an improved spiking silicon neuron (SN) [6] that approximates the dynamics of ionic currents of a real nerve cell. The improved version has less circuitry and fewer parameters than previous circuits thereby improving the spiking characteristics. We describe the differential equations governing the revised circuits and use them to explai...

Retention of parameters and learnt synaptic weights is a central problem in the construction of neural networks. We have applied analog floating gate technology to solve these problems in the context of biologically realistic `silicon neurons'. Parameters are stored on a novel floating gate array, and synaptic weights are retained by a floating gat...

We have developed compact analog integrated circuits that simulate two synaptic excitatory conductances. A four-transistor circuit captures the dynamics of an excitatory postsynaptic current caused by a real AMPA conductance. A six-transistor circuit simulates the effects of a real voltage-dependent NMDA conductance. The postsynaptic current dynami...

this paper, we describe how the silicon neuron can be

Spontaneous fixations onto shapes are driven by a structural analysis (e.g. Richard and Kaufman 1969; Melcher and Kowler, 1999). But is such analysis also carried out during free viewing of real-world scenes? Here, we analyze how fixation locations in such scenes are related to their region using their symmetric axes as a reference. Each fixation l...

The idea of gaze guidance is to lead a viewer's gaze through a visual display in order to facilitate his/her search for specific information. This study elaborates on the process of guiding gaze from one spatial position to another, whereby the goal is to create a guidance process that is as least-obtrusive as possible. A list of guidance aspects i...

Research on gaze control has been typically characterized for stationary targets on homogenous backgrounds. In this study, gaze behavior is investigated with a dynamic broadband (1/f) noise display, which offers much more the kind of natural background distraction as when moving through natural scenes. To elucidate the possible bottom-up component...

Visual orienting has been typically characterized using simple displays, e.g. displays with a static target placed on a homogenous background. In this study, visual orienting was investigated using a dynamic broadband (1/f) noise display that should mimic a more naturalistic setting and that should allow performing saccadic orienting experiments wi...

How structure is possibly represented by the visual system is still little understood. A promising concept is the symmetric-axis transform, which has already been implemented for the purpose of shape identification. However, no instantiation of this transform has been provided yet, that generates the symmetric axes for open contours. This can be ac...