Oriented Line Segments Research Articles

EMS: 3D Eyebrow Modeling from Single-View Images

Eyebrows play a critical role in facial expression and appearance. Although the 3D digitization of faces is well explored, less attention has been drawn to 3D eyebrow modeling. In this work, we propose EMS, the first learning-based framework for single-view 3D eyebrow reconstruction. Following the methods of scalp hair reconstruction, we also represent the eyebrow as a set of fiber curves and convert the reconstruction to fibers growing problem. Three modules are then carefully designed: RootFinder firstly localizes the fiber root positions which indicate where to grow; OriPredictor predicts an orientation field in the 3D space to guide the growing of fibers; FiberEnder is designed to determine when to stop the growth of each fiber. Our OriPredictor directly borrows the method used in hair reconstruction. Considering the differences between hair and eyebrows, both RootFinder and FiberEnder are newly proposed. Specifically, to cope with the challenge that the root location is severely occluded, we formulate root localization as a density map estimation task. Given the predicted density map, a density-based clustering method is further used for finding the roots. For each fiber, the growth starts from the root point and moves step by step until the ending, where each step is defined as an oriented line segment with a constant length according to the predicted orientation field. To determine when to end, a pixel-aligned RNN architecture is designed to form a binary classifier, which outputs stop or not for each growing step. To support the training of all proposed networks, we build the first 3D synthetic eyebrow dataset that contains 400 high-quality eyebrow models manually created by artists. Extensive experiments have demonstrated the effectiveness of the proposed EMS pipeline on a variety of different eyebrow styles and lengths, ranging from short and sparse to long bushy eyebrows.

Two-photon imaging of V1 responses to “pop-out” texture patterns in awake macaques

A line segment among an array of uniformly differently oriented line segments can be detected effortlessly. Various experimental and modeling studies have linked this ‘pop-out’ phenomenon to V1 processing. Here we studied the V1 neuronal activities associated with visual “pop-out” in V1 superficial- layer neurons in awake macaques with two-photon calcium imaging (GCaMP5). The target stimulus was a drifting target Gabor (2-cycles/sec speed, 90%-contrast, 6-cpd SF, 2o- eccentricity) at various orientations. The surrounding stimuli were 1-2 rings of Gabors with uniform or random orientations around the center stimulus. Neuronal responses from 1119 orientation-tuned cells 150 μm deep from the cortical surface were recorded. When the target was surrounded by one ring of uniform Gabors, the cells’ optimal responses were suppressed, with stronger suppression at collinear orientation, and weaker suppression as surround orientation deviated from collinear, which floored at 45o deviation. Adding one or two more rings of Gabors at collinear orientation increased surround suppression, reducing the median R_sur/R_base ratio from 0.578 to 0.496 and 0.462. This trend persisted with surround stimulus orientation up to 22.5o deviation from collinear. In contrast, at 45o or larger deviation, increasing the number of surround rings reduced surround suppression. At orthogonal surround orientation, the R_sur/R_base ratio increased from 0.658 to 0.736 and 0.758. In addition, surround suppression by randomly oriented increased with more rings of surround stimuli, as in the case of collinear surround stimuli. These results demonstrate significantly lower surround suppression when a target is imbedded in orthogonal than in collinear surround stimuli, especially with the surround stimulus set is large and resembles “pop-out” textures (R_sur/R_base ratio = 0.758. Vs. 0.462 with 3 rings of orthogonal vs. collinear stimuli). In “pop-out” textures, far orthogonal surround stimuli may disinhibit near surround stimuli to relieve surround suppression and achieve a pop out percept.

Usefulness of cutting planes in the hierarchical segmentation of cardiac anatomical structures

A spatial geometric plane is defined by the three-dimensional coordinates of a pair of spatial points and the direction that the normal vector establishes, which is formed by joining those points by means of an oriented line segment. This type of planes, in three-dimensional images, is extremely useful as an alternative solution to the problem of low contrast that exhibit the anatomical structures present in cardiac computed tomography images. To do this, after using a predetermined filter bank and in order to define a region of interest, a smart operator based on least squares support vector machines is trained and validated in order to detect the aforementioned coordinates which enables the location of the plane, in the three-dimensional space that contains the considered images. Once the structure that is required to segment is identified, a discriminant function is used that cancels all information not linked to this structure. In this work, the segmentation of the left ventricle, based on region growing technique, is firstly considered and then the left atrium is segmented considering region growing technique and an inverse discriminant function. The results show an excellent correspondence relationship when the spatial union of both structures is made.

Determination of the Wave Parameters from the Statistical Characteristics of the Image of a Linear Test Object

We statistically analyze the images of the objects of the “light-line” and “half-plane” types which are observed through a randomly irregular air-water interface. The expressions for the correlation function of fluctuations of the image of an object given in the form of a luminous half-plane are found. The possibility of determining the spatial and temporal correlation functions of the slopes of a rough water surface from these relationships is shown. The problem of the probability of intersection of a small arbitrarily oriented line segment by the contour image of a luminous straight line is solved. Using the results of solving this problem, we show the possibility of determining the values of the curvature variances of a rough water surface. A practical method for obtaining an image of a rectilinear luminous object in the light rays reflected from the rough surface is proposed. It is theoretically shown that such an object can be synthesized by temporal accumulation of the image of a point source of light rapidly moving in the horizontal plane with respect to the water surface.

Contour Integration in Dynamic Scenes: Impaired Detection Performance in Extended Presentations.

Since scenes in nature are highly dynamic, perception requires an on-going and robust integration of local information into global representations. In vision, contour integration (CI) is one of these tasks, and it is performed by our brain in a seemingly effortless manner. Following the rule of good continuation, oriented line segments are linked into contour percepts, thus supporting important visual computations such as the detection of object boundaries. This process has been studied almost exclusively using static stimuli, raising the question of whether the observed robustness and “pop-out” quality of CI carries over to dynamic scenes. We investigate contour detection in dynamic stimuli where targets appear at random times by Gabor elements aligning themselves to form contours. In briefly presented displays (230 ms), a situation comparable to classical paradigms in CI, performance is about 87%. Surprisingly, we find that detection performance decreases to 67% in extended presentations (about 1.9–3.8 s) for the same target stimuli. In order to observe the same reduction with briefly presented stimuli, presentation time has to be drastically decreased to intervals as short as 50 ms. Cueing a specific contour position or shape helps in partially compensating this deterioration, and only in extended presentations combining a location and a shape cue was more efficient than providing a single cue. Our findings challenge the notion of CI as a mainly stimulus-driven process leading to pop-out percepts, indicating that top-down processes play a much larger role in supporting fundamental integration processes in dynamic scenes than previously thought.

Recovering metric properties of objects through spatiotemporal interpolation.

Spatiotemporal interpolation (STI) refers to perception of complete objects from fragmentary information across gaps in both space and time. It differs from static interpolation in that requirements for interpolation are not met in any static frame. It has been found that STI produced objective performance advantages in a shape discrimination paradigm for both illusory and occluded objects when contours met conditions of spatiotemporal relatability. Here we report psychophysical studies testing whether spatiotemporal interpolation allows recovery of metric properties of objects. Observers viewed virtual triangles specified only by sequential partial occlusions of background elements by their vertices (the STI condition) and made forced choice judgments of the object's size relative to a reference standard. We found that length could often be accurately recovered for conditions where fragments were relatable and formed illusory triangles. In the first control condition, three moving dots located at the vertices provided the same spatial and timing information as the virtual object in the STI condition but did not induce perception of interpolated contours or a coherent object. In the second control condition oriented line segments were added to the dots and mid-points between the dots in a way that did not induce perception of interpolated contours. Control stimuli did not lead to accurate size judgments. We conclude that spatiotemporal interpolation can produce representations, from fragmentary information, of metric properties in addition to shape.

Colour-specific differences in attentional deployment for equiluminant pop-out colours: Evidence from lateralised potentials

We investigated how target colour affected behavioural and electrophysiological results in a visual search task. Perceptual and attentional mechanisms were tracked using the N2pc component of the event-related potential and other lateralised components. Four colours (red, green, blue, or yellow) were calibrated for each participant for luminance through heterochromatic flicker photometry and equated to the luminance of grey distracters. Each visual display contained 10 circles, 1 colored and 9 grey, each of which contained an oriented line segment. The task required deploying attention to the colored circle, which was either in the left or right visual hemifield. Three lateralised ERP components relative to the side of the lateral coloured circle were examined: a posterior contralateral positivity (Ppc) prior to N2pc, the N2pc, reflecting the deployment of visual spatial attention, and a temporal and contralateral positivity (Ptc) following N2pc. Red or blue stimuli, as compared to green or yellow, had an earlier N2pc. Both the Ppc and Ptc had higher amplitudes to red stimuli, suggesting particular selectivity for red. The results suggest that attention may be deployed to red and blue more quickly than to other colours and suggests special caution when designing ERP experiments involving stimuli in different colours, even when all colours are equiluminant.

Closure enhancement in a model network with orientation tuned long-range connectivity

The primary visual cortex (V1) of the mammalian brain is equipped with a specifically connected network of neurons that can potentially solve difficult image processing tasks. These neurons are selectively tuned for locations in visual space and also for line orientation. The coupling of location and orientation tuning results in the neural representation of the visual world in terms of local features. These local features, e.g., oriented line segments, will have to be linked together in order to parse the visual world into regions corresponding to object and ground. Although standard models of V1 do not address the issue of interacting neuronal populations, we suggest that the long-range connectivity pattern of V1 provides an architecture where spreading neural activity may lead to pertinent figure-ground segmentation. The model relies on the fact that in addition to the processing units, their connections are also selectively tuned for space and orientation. From the computational point of view, the mod...

Top-Down Control in Contour Grouping

Human observers tend to group oriented line segments into full contours if they follow the Gestalt rule of 'good continuation'. It is commonly assumed that contour grouping emerges automatically in early visual cortex. In contrast, recent work in animal models suggests that contour grouping requires learning and thus involves top-down control from higher brain structures. Here we explore mechanisms of top-down control in perceptual grouping by investigating synchronicity within EEG oscillations. Human participants saw two micro-Gabor arrays in a random order, with the task to indicate whether the first (S1) or the second stimulus (S2) contained a contour of collinearly aligned elements. Contour compared to non-contour S1 produced a larger posterior post-stimulus beta power (15–21 Hz). Contour S2 was associated with a pre-stimulus decrease in posterior alpha power (11–12 Hz) and in fronto-posterior theta (4–5 Hz) phase couplings, but not with a post-stimulus increase in beta power. The results indicate that subjects used prior knowledge from S1 processing for S2 contour grouping. Expanding previous work on theta oscillations, we propose that long-range theta synchrony shapes neural responses to perceptual groupings regulating lateral inhibition in early visual cortex.

Approximation algorithms for deployment of sensors for line segment coverage in wireless sensor networks

The coverage problem in wireless sensor networks deals with the problem of covering a region or parts of it with sensors. In this paper, we address the problem of covering a set of line segments with minimum number of sensors. A line segment ℓ is said to be 1-covered if it intersects the sensing region of at least one among the sensors distributed in a bounded rectangular region R. We assume that the sensing radius of each sensor is uniform. The problem of finding the minimum number of sensors needed to 1-cover each member in a given set of line segments in R is NP-hard. We propose two constant factor approximation algorithms and a PTAS (polynomial time approximation scheme) for the problem for 1-covering a set of axis-parallel line segments. We also show that a PTAS exists for 1-covering a set of arbitrarily oriented line segments in R where the lengths of the line segments are bounded within a constant factor of the sensing radius of each sensor. Finally, we propose a constant factor approximation algorithm for k-covering axis-parallel line segments such that sensors maintain a minimum separation among them.

Temporal resolution of orientation-defined texture segregation: a VEP study

Orientation is one of the visual dimensions that subserve figure-ground discrimination. A spatial gradient in orientation leads to "texture segregation", which is thought to be concurrent parallel processing across the visual field, without scanning. In the visual-evoked potential (VEP) a component can be isolated which is related to texture segregation ("tsVEP"). Our objective was to evaluate the temporal frequency dependence of the tsVEP to compare processing speed of low-level features (e.g., orientation, using the VEP, here denoted llVEP) with texture segregation because of a recent literature controversy in that regard. Visual-evoked potentials (VEPs) were recorded in seven normal adults. Oriented line segments of 0.1 degrees x 0.8 degrees at 100% contrast were presented in four different arrangements: either oriented in parallel for two homogeneous stimuli (from which were obtained the low-level VEP (llVEP)) or with a 90 degrees orientation gradient for two textured ones (from which were obtained the texture VEP). The orientation texture condition was presented at eight different temporal frequencies ranging from 7.5 to 45 Hz. Fourier analysis was used to isolate low-level components at the pattern-change frequency and texture-segregation components at half that frequency. For all subjects, there was lower high-cutoff frequency for tsVEP than for llVEPs, on average 12 Hz vs. 17 Hz (P = 0.017). The results suggest that the processing of feature gradients to extract texture segregation requires additional processing time, resulting in a lower fusion frequency.

On intersecting a set of parallel line segments with a convex polygon of minimum area

Let S = {l1, l2, l3, . . . , ln} be a set of n vertical line segments in the plane. Though not essential, to simplify proofs we assume that no two li ’s are on the same vertical line. A convex polygon weakly intersects S if it contains a point of each line segment on its boundary or interior. In this paper, we propose an O(n logn) algorithm for the problem of finding a minimum area convex polygon that weakly intersects S. The principal motivation behind this paper is the open problem proposed by Tamir [5] at the fourth Computational Geometry day at NYU to decide if there exists a convex polygon whose boundary intersects a set of arbitrarily oriented line segments.

Data Acceptance for Automated Leukocyte Tracking Through Segmentation of Spatiotemporal Images

A crucial task in inflammation research and inflammatory drug validation is leukocyte velocity data collection from microscopic video imagery. Since manual methods are bias-prone and extremely time consuming, automated tracking methods are required to compute cell velocities. However, an automated tracking method is of little practical use unless it is accompanied by a mechanism to validate the tracker output. In this paper, we propose a validation technique that accepts or rejects the output of automated tracking methods. The proposed method first generates a spatiotemporal image from the cell locations given by a tracking method; then, it segments the spatiotemporal image to detect the presence or absence of a leukocyte. For segmenting the spatiotemporal images, we employ an edge-direction sensitive nonlinear filter followed by an active contour based technique. The proposed nonlinear filter, the maximum absolute average directional derivative (MAADD), first computes the magnitude of the mean directional derivative over an oriented line segment and then chooses the maximum of all such values within a range of orientations of the line segment. The proposed active contour segmentation is obtained via growing contours controlled by a two-dimensional force field, which is constructed by imposing a Dirichlet boundary condition on the gradient vector flow (GVF) field equations. The performance of the proposed validation method is reported here for the outputs of three different tracking techniques: the method was successful in 97% of the trials using manual tracking, in 94% using correlation tracking and in 93% using active contour tracking.

Response Modulations by Static Texture Surround in Area V1 of the Macaque Monkey Do Not Depend on Feedback Connections From V2

We analyzed the extracellular responses of 70 V1 neurons (recorded in 3 anesthetized macaque monkeys) to a single oriented line segment (or bar) placed within the cell classical receptive field (RF), or center of the RF. These responses could be modulated when rings of bars were placed entirely outside, but around the RF (the "near" surround region), as described in previous studies. Suppression was the main effect. The response was enhanced for 12 neurons when orthogonal bars in the surround were presented instead of bars having the same orientation as the center bar. This orientation contrast property is possibly involved in the mediation of perceptual pop-out. The enhancement was delayed compared with the onset of the response by about 40 ms. We also observed a suppression originating specifically from the flanks of the surround. This "side-inhibition," significant for nine neurons, was delayed by about 20 ms. We tested whether these center/surround interactions in V1 depend on feedback connections from area V2. V2 was inactivated by GABA injections. We used devices made of six micropipettes to inactivate the convergent zone from V2 to V1. We could reliably inactivate a 2- to 4-mm-wide region of V2. Inactivation of V2 had no effect on the center/surround interactions of V1 neurons, even those that were delayed. Therefore the center/surround interactions of V1 neurons that might be involved in pop-out do not appear to depend on feedback connections from V2, at least in the anesthetized monkey. We conclude that these properties are probably shaped by long-range connections within V1 or depend on other feedback connections. The main effect of V2 inactivation was a decrease of the response to the single bar for about 10% of V1 neurons. The decrease was delayed by <20 ms after the response onset. Even the earliest neurons to respond could be affected by the feedback from V2. Together with the results on feedback connections from MT (previous paper), these findings show that feedback connections potentiate the responses to stimulation of the RF center and are recruited very early for the treatment of visual information.

Long range interactions between oriented texture elements.

Long range interactions between texture elements (short, oriented line segments) were examined. Specifically, we studied the influence of a background array of texture elements on the detectability of a target element (separated from the background by an intermediate textured region) using textures like those of Caputo (Vis. Res. 1996, 36, 2815-2826). We found that, in general, when the background elements were oriented orthogonally to the target element, detection of the target element was better than when the background elements had the same orientation as the target element. We discuss these interactions in terms of inhibitory and excitatory connections between orientation and spatial frequency selective linear filters (e.g. filters which mimic V1 simple cells) which would respond to the individual texture elements.

Examining edge- and region-based texture analysis mechanisms.

Instantaneous texture discrimination performance was examined for different texture stimuli to uncover the use of edge-based and region-based texture analysis mechanisms. Textures were composed of randomly placed, short, oriented line segments. Line segment orientation was chosen randomly using a Gaussian distribution (described by a mean and a standard deviation). One such distribution determined the orientations on the left side of the image, and a second distribution was used for the right side. The two textures either abutted to form an edge or were separated by a blank region. A texture difference in mean orientation led to superior discrimination performance when the textures abutted. On the other hand, when the textures differed in the standard deviation of the orientation distribution, performance was similar in the two conditions. These results suggest that edge-based texture analysis mechanisms were used (i.e. were the most sensitive) in the abutting difference-in-mean case, but region-based texture analysis mechanisms were used in the other three cases.

Collinearity and Parallelism are Statistically Significant Second-Order Relations of Complex Cell Responses

By investigating the second-order statistics of Gabor wavelet responses derived from natural images, we show that collinearity and parallelism are conspicuous relations. We give a precise mathematical characterization of these Gestalt principles by the conditional probability of two responses. Essential for our investigations is a non-linear transformation, initially utilized within the object recognition system [5], which transforms continuous Gabor wavelet responses into a binary code indicating the presence or absence of local oriented line segments.

Anisotropy in Oriented-Line-Target Detection at High and Low Luminance Contrast

Observers can detect a uniquely oriented line segment (‘target’) in a background field of uniformly oriented line segments (‘nontargets’) even if viewing time is brief. When the lines have high luminance contrast, the variation of orientation increment threshold with nontarget orientation is periodic, generally with a period of about 90° although smaller periods have been found. Do the orientation-sensitive mechanisms giving rise to periodicities function only at high contrast? This question was addressed in a line-target detection experiment. Twenty white line segments of length 1 deg visual angle were presented in a circular field of diameter 20 deg visual angle. Nontarget orientations were in the range 0°, 5°, …, 175°, and the difference between nontarget and target orientations was varied adaptively. Stimulus displays lasting 40 ms were followed by a blank interstimulus interval lasting 60 ms and then a random-line mask. A view-tunnel provided a grey background of luminance 35 cd m−2 and stimulus contrast was 0.1, 0.3, 0.5, 0.7, or 0.9 log unit above the observer's luminance detection threshold for one line segment. When contrast was 0.1 log unit above this threshold, performance was near chance level. As contrast increased from 0.3 to 0.9 log unit above luminance detection threshold, performance improved and orientation increment thresholds decreased, showing that early orientation-processing is most effective at high luminance contrast. Nonetheless, periodicities were found in all conditions where performance was better than chance which suggests that the orientation-sensitive mechanisms associated with periodicities operate at both high and low luminance contrast.

Basic visual capacities and shape discrimination after lesions of extrastriate area V4 in macaques.

Ibotenic acid lesions were made in four macaque monkeys in a region of cortical area V4 that corresponds to the lower quadrant of one hemifield. For visual testing, fixation locus was monitored with scleral search coils and controlled behaviorally to place test stimuli either in the lesioned quadrant or in a control location in the opposite hemifield. Some basic visual capacities were slightly altered by the lesions; there was a two-fold reduction of luminance contrast sensitivity as well as red-green chromatic contrast sensitivity, both tested with stationary gratings. On the other hand, little or no loss was found when contrast sensitivity for detection or direction discrimination was tested with 10-Hz drifting gratings nor was there a reliable change in visual acuity. Hue and luminance matching were tested with a spatially more complex matching-to-sample task, but monkeys could not learn this task in the visual field locus of a V4 lesion. If previously trained at this locus, performance was not affected by the lesion. In contrast to the small effects on basic visual capabilities, performance on two form discrimination tasks was devastated by V4 lesions. The first involved discriminating the orientation of colinear groups of dots on a background of randomly placed dots. The second involved discriminating the orientation of a group of three line segments surrounded by differently oriented line segments. Some selectivity of the deficits for form discrimination was shown by the lack of an effect of the lesions on a global motion discrimination. These results show that while V4 lesions cause only slight disruptions of basic visual capacities, they profoundly disrupt form discriminations.

Multiple-element line segment precues: orientation and location effects on attention.

Three experiments were conducted to examine the hypothesis that a uniquely oriented line segment located at the end of the directional flow produced by the alignment of the long axes of identically oriented background segments will not be detected as rapidly as when the segment appears at locations within the flow. The hypothesis was supported with response time measured in Experiment 1. A location-cuing task was used in the last two experiments, with the unique segment becoming a precue to indicate the location of an upcoming target character, accuracy was the dependent measure. Poorer detection of the unique segment at end-of-flow locations was shown not to be due to a local configuration effect produced by the uniquely oriented segment in conjunction with the segment on either side. The time course of attentional development with multiple-segment precue displays was comparable to that with an arrow at fixation as the precue, and slower than that with a line segment appearing alone.