WSCG '2016: Short Papers Proceedings

Permanent URI for this collection

http://hdl.handle.net/11025/26401

Browse

Showing 1 - 20 out of 46 results

Face tracking using a combination of colour and pattern matching based on particle filter
(Václav Skala - UNION Agency, 2016) Gonzalez, Matilde; Collet, Christophe; Skala, Václav
Robust real-time face tracking is an important and challenging task in computer vision applications. In this paper, we propose a novel particle filter algorithm to robustly track faces. Particle observations are computed by considering cue and appearance feature. Cue feature is used to identify skin regions, e.i. face and hands, while appearance is used to directly label targets. Normalized Cross Correlation (NCC) between an image template and particle samples is computed to robustly find the face among other skin regions. In other words, the image template is registered to a frame using particle filter to perform the optimization. Real-time is achieved by using integral images to compute image features. Evaluation results show the advantages and limitation of our approach.
Handwritten digit recognition by support vector machine optimized by Bat algorithm
(Václav Skala - UNION Agency, 2016) Tuba, Eva; Tuba, Milan; Simian, Dana; Skala, Václav
Handwritten digit recognition is an important but very hard practical problem. This is a classification problem for which support vector machines are very successfully used. Determining optimal support vector machine is another hard optimization problem that involves tuning of the soft margin and kernel function parameters. For this optimization we adjusted recent swarm intelligence bat algorithm. We intentionally used weak set of features, four histogram projections, to prove that even under unfavorable conditions our algorithm would achieve acceptable results. We tested our approach on standard MNIST benchmark datasets and compared the results with other recent approaches from literature where our proposed algorithm achieved better results i.e. higher correct classification percentage.
Integration of statistical spatial relations into Active Shape Model- Application to striatum segmentation in MRI
(Václav Skala - UNION Agency, 2016) Ettaïeb, Saïd; Mnassri, Besma; Hamrouni, Kamel; Skala, Václav
This paper describes a new method based on Active Shape Model (ASM) and statistical spatial relations. It combines three types of a priori knowledge: the structures shapes, the distance and the angle variability between them. This knowledge is estimated during a training step. Then, the obtained models are used to guide the evolution of initial shapes during the segmentation step. The proposed method is applied to extract the striatum (Caudate nucleus and Putamen) on MR images of the brain. The obtained results are promising and show the performance of the proposed method.
A novel accurate 3D surfaces description using the Arc-length reparametrized level curves of the three-polar representation
(Václav Skala - UNION Agency, 2016) Rihani, Amal; Jribi, Majdi; Ghorbel, Faouzi; Skala, Václav
This paper studies the problem of the 3D surfaces representation. Our starting point is the extraction of the threepolar representation from the 3D shapes. It consists on a level curves set of the superposition of the three geodesic potentials generated from three reference points of the surface. These curves are characterized by their invariance under the M(3) group of R3 displacements. We intend to make the arc-length reparametrization of each level curve to ensure its independence to the initial parametrization. The novel representation is materialized by the points of the arc-length reparametrization of all the level curves. Therefore, we obtain an invariant representation under the M(3) transformations group and independent to the initial parametrization. In this work, we implement it on 3D faces since this type of surfaces knows actually a growing interest for the identities determination especially after the many terrorist acts occurred around the world. We experiment, in this context, the identification scenario on a part of the BU-3DFE database. The obtained results show the accuracy of the novel representation.
Automatic segmentation of cervical cells in Pap smear images
(Václav Skala - UNION Agency, 2016) Boughzala, Omelkhir; Guesmi, Lamia; Ben Abdallah, Asma; Hédi Bedoui, Mohamed; Skala, Václav
In the context of medical diagnosis by image analysis, segmentation is the most critical step in image processing. The problem of image segmentation has been studied for years and many methods have been suggested in the literature. However, there is not yet any automatic method able to correctly process any type of image. In this work, we present an automated method for cell segmentation in Pap smear images. The automatic analysis of Pap smear images is one of the most interesting fields in medical image processing. The object of this paper is to present the strategy of the first part of the system segmentation. It is based on a segmentation of color images tested with different classical color spaces, namely RGB, L*a*b, HSV, and YCbCr, to select the best color space using k-means clustering to separate groups of objects. The k means clustering treats each object as having a location in space. The method is aimed at developing an automated Pap smear analysis system which can help cytotechnologists reduce examination time in pap screening process.
Fast and robust construction of 3D architectural models from 2D plans
(Václav Skala - UNION Agency, 2016) Pandey, Jalaj; Sharma, Ojaswa; Skala, Václav
In this work we present a simple and robust method to create 3D building models from a set of architectural plans. Such plans are created for human readability and thus pose some problem in automatic creation of a 3D model. We suggest a semi-automated approach for plan cleaning and provide an algorithm for alignment and stacking of the plans followed by generation of 3D building model. We show results of our method on floor plans that generate complex 3D models in near real-time.
An unsupervised 3D mesh segmentation based on HMRF-EM algorithm
(Václav Skala - UNION Agency, 2016) Mabrouk, Sabra; Chaieb, Faten; Ghorbel, Faouzi; Skala, Václav
We propose a new 3D mesh segmentation method based on the HMRF-EM framework. The clustering method relies on the curvature attribute and considers the spatial information encoded by the mutual influences of neighboring mesh elements. A region growing process is then carried out in order to extract connected regions followed by a merging procedure. The purpose of this latter process is to only preserve meaningful regions. Experiments conducted on different meshes are encouraging and show that the proposed method gives satisfying results compared with classical statistical ones such as kmeans and EM algorithms.
A new approach to turbid water surface identification for autonomous navigation
(Václav Skala - UNION Agency, 2016) Colet, Mateus Eugênio; Braun, Adriana; Manssour, Isabel H.; Skala, Václav
Navigation of autonomous vehicles in natural environments based on image processing is certainly a complex problem due to the dynamic characteristics of aquatic surfaces, such as brightness and color saturation. This paper presents a new approach to identify turbid water surfaces based on their optical properties, aiming to allow automatic navigation of autonomous vehicles regarding inspection, mitigation and management of aquatic natural disasters. More specifically, computer vision techniques were employed in conjunction to artificial neural networks (ANNs), in order to build a classifier designed to generate a navigation map that is interpreted by a state machine for decision making. To do so, a study on the use of different features based on color and texture of such turbid surfaces was conducted. In order to compress the extracted information, Principal Component Analysis (PCA) was performed and its results were used as inputs to ANN. The whole developed approach was embedded in an aquatic vehicle, and results and assessments were validated in real environments and different scenarios.
Facial expression recognition using salient facial patches
(Václav Skala - UNION Agency, 2016) Mliki, Hazar; Hammami, Mohamed; Skala, Václav
This paper proposes a novel facial expression recognition method composed of two main steps: offline step and online step. The offline step selects the most salient facial patches using mutual information technique. The online step relies on the already selected patches to identify the facial expression using an SVM classifier. In both steps, the LBP operator was used to extract facial expressions features. Through an extensive experiments on the JAFFE and KANADE databases, we have shown that our method, thanks to the salient selected patches, has the advantage of being much faster with a significant gain in recognition performance.
MorphableUI: a hypergraph-based approach to distributed multimodal interaction for rapid prototyping and changing environments
(Václav Skala - UNION Agency, 2016) Krekhov, Andrey; Grüninger, Jürgen; Baum, Kevin; McCann, David; Krüger, Jens; Skala, Václav
Nowadays, users interact with applications in constantly changing environments. The plethora of I/O modalities is beneficial for a wide range of application areas such as virtual reality, cloud-based software, or scientific visualization. These areas require interfaces based not only on the traditional mouse and keyboard but also on gestures, speech, or highly-specialized and environment-dependent equipment. We introduce a hypergraph-based interaction model and its implementation as a distributed system, called MorphableUI. Its primary focus is to deliver a user- and developer-friendly way to establish dynamic connections between applications and interaction devices. We present an easy-to-use API for developers and a mobile frontend for users to set up their preferred interfaces. During runtime, MorphableUI transports interaction data between devices and applications. As one of the novelties, the system supports I/O transfer functions by automatically splitting, merging, and casting inputs from different modalities. MorphableUI emphasizes rapid prototyping and, e.g., facilitates the execution of user studies due to easy UI reconfiguration and device exchangeability.
Kinematical ruled surfaces based on interrelated movements in triads of contacted axoids
(Václav Skala - UNION Agency, 2016) Rachkovskaya, Galina S.; Kharabayev, Yuriy N.; Rachkovskaya, Natalya S.; Skala, Václav
Kinematical ruled surfaces are constructed by generating the line's motion of a moving ruled surface during its movement along a fixed ruled surface [Spr02a]. The main condition of constructing kinematical ruled surfaces is that a moving axoid contacts with a fixed axoid along their common generating line in each of their positions during the movement of one axoid along another. A lot of well-known kinematical ruled surfaces are constructed on the base of certain pairs of contacted axoids such as “plane – cylinder”, “plane – cone”, “cylinder – cylinder”, “cone – cone”, etc. [Kri06a]. A new model of constructing kinematical ruled surfaces based on interrelated movements in the triads of contacted axoids is proposed in this research. Geometrical models, analytical representations, and computer visualization of the new constructed kinematical surfaces for some cases of triads of contacted axoids “plane – cylinder – cylinder”, “plane – cone – cone”, “cylinder – cylinder – cylinder”, “cone – cone – cone” (Fig. 1), and for matched triads of one-sheet hyperboloids of revolution are developed in this paper. Figures of the triads of contacted axoids and corresponding constructed kinematical ruled surfaces have been developed with the help of the software application AMG (“ArtMathGraph”) [Con07a].
Measuring event probabilities in uncertain scalar datasets using Gaussian processes
(Václav Skala - UNION Agency, 2016) Schlegel, Steven; Volke, Sebastian; Scheuermann, Gerik; Skala, Václav
In this paper, we show how the concept of Gaussian process regression can be used to determine potential events in scalar data sets. As a showcase, we will investigate climate data sets in order to identify potential extrem weather events by deriving the probabilities of their appearances. The method is implemented directly on the GPU to ensure interactive frame rates and pixel precise visualizations. We will see, that this approach is especially well suited for sparse sampled data because of its reconstruction properties.
Flexible calibration of color and depth camera arrays
(Václav Skala - UNION Agency, 2016) Avetisyan, Razmik; Rosenke, Christian; Staadt, Oliver; Skala, Václav
In this work we present a flexible approach for calibrating an array of multiple stationary color and depth cameras using an optical tracking system. Our application domain is focused on 3D telepresence. Calibrating cameras in this area is still a major problem due to the limited applicability of common calibration approaches. Usually, groups of cameras are calibrated relative to each other by either requiring heavily overlapping fields of view for many pairs of participating cameras or free movable cameras. Our method moves away from these techniques by calibrating every camera individually. The key technology applied is a tracked calibration target with permanently identified global location provided by a tracking device. Detecting the known target geometry in a camera image provides, beside intrinsic calibration parameters, the position of the camera relative to the target. Combining these two aspects of the calibration target’s location makes it possible to register every camera in the common tracking coordinate system. We validate our approach using our prototype with 12 Firewire color cameras, 3 Kinect depth cameras, an OptiTrack tracking device, and a checkerboard with an attached trackable rigid body (see Figure 1). In this setup, we achieve a reprojection error of below 0:5 pixels on average.
Temporal filtering of depth Images using optical flow
(Václav Skala - UNION Agency, 2016) Avetisyan, Razmik; Rosenke, Christian; Luboschik, Martin; Staadt, Oliver; Skala, Václav
We present a novel depth image enhancement approach for RGB-D cameras such as the Kinect. Our approach employs optical flow of color images for refining the quality of corresponding depth images. We track every depth pixel over a sequence of frames in the temporal domain and use valid depth values of the same point for recovering missing and inaccurate information. We conduct experiments on different test datasets and present visually appealing results. Our method significantly reduces the temporal noise level and the flickering artifacts.
Toward a computational model and decision support system for reducing errors in pharmaceutical packaging
(Václav Skala - UNION Agency, 2016) Quigley, Carson; Shooter, Steven; Mitchel, Aaron; Skala, Václav
The US Institute of Medicine reports that one medication error occurs per patient per day in hospital care, and other studies indicate that medication administration errors attributed to packaging and/or labeling confusion can be as high as 33%. While many engineered products have identifiable features that help establish commonality and differentiation within a product family, vital features of consumable products such as medications are often not readily apparent in their physical form. As a result, caregivers must rely on the labeling and packaging to effectively determine the contents. Adverse Drug Events (ADEs) are the most common category of medical errors and include wrong drug, wrong dose, wrong route of administration, and wrong patient. It is estimated that in the US each year, medication errors harm at least 1.5 million people, resulting in 106,000 deaths. Computational models and associated decision support systems have the potential to improve pharmaceutical delivery safety through informed design of packaging features and enhanced situational awareness and decision-making during drug identification and administration. Past research has led to the formulation of measures for representing the degree of commonality and differentiation of packaging features in pharmaceutical families or versus look-alike drugs. Preliminary studies have validated these measures of feature prominence based on feature size and location. This paper describes a study using eye tracking to evaluate gaze patterns and further validate these measures. The results support the measures and indicate that increased commonality of features results in shorter reaction times, but also shorter fixation times. These results have implications in the formulation of a resulting decision support system.
Background modeling using perception-based local pattern
(Václav Skala - UNION Agency, 2016) Chan, K. L.; Skala, Václav
Background modeling is an important issue in video surveillance. A sophisticated and adaptive background model can be used to detect moving objects which are segregated from the scene in each image frame of the video via the background subtraction process. Many background subtraction methods are proposed for video acquired by a stationary camera, assuming that the background exhibits stationary properties. However, it becomes harder under various dynamic circumstances – illumination changes, background motions, shadows, camera jitter, etc. We propose a versatile background modeling method for representing complex background scenes. The background model is learned from a short sequence of spatio-temporal video data. Each pixel of the background scene is represented by samples of color and local pattern. The local pattern is characterized by perception-inspired features. In order to cater for changes in the scene, the background model is updated along the video based on the background subtraction result. In each new video frame, moving objects are considered as foregrounds which are detected by background subtraction. A pixel is labeled as background when it matches with some samples in the background model. Otherwise, the pixel is labeled as foreground. We propose a novel perception-based matching scheme to estimate the similarity between the pixel and the background model. We test our method using common datasets and achieve better performance than various background subtraction algorithms in some image sequences.
Integrating depth-HOG and spatio-temporal joints data for action recognition
(Václav Skala - UNION Agency, 2016) Arora, Noopur; Shukla, Parul; Biswas, Kanad K.; Skala, Václav
In this paper, we propose an approach for human activity recognition using gradient orientation of depth maps and spatio-temporal features from body-joints data. Our approach is based on an amalgamation of key local and global feature descriptors such as spatial pose, temporal variation in ‘joints’ position and spatio-temporal gradient orientation of depth maps. Additionally, we obtain a motion-induced global shape feature describing the motion dynamics during an action. Feature selection is carried out to select a relevant subset of features for action recognition. The resultant features are evaluated using SVM classifier. We validate our proposed method on our own dataset consisting of 11 classes and a total of 287 videos. We also compare the effectiveness of our method on the MSR-Action3D dataset.
A User-centered approach for optimizing information visualizations
(Václav Skala - UNION Agency, 2016) Baum, David; Kovacs, Pascal; Eisenecker, Ulrich; Müller, Richard; Skala, Václav
The optimization of information visualizations is time consuming and expensive. To reduce this we propose an improvement of existing optimization approaches based on user-centered design, focusing on readability, comprehensibility, and user satisfaction as optimization goals. The changes comprise (1) a separate optimization of user interface and representation, (2) a fully automated evaluation of the representation, and (3) qualitative user studies for simultaneously creating and evaluating interface variants. On the basis of these results we are able to find a local optimum of an information visualization in an efficient way.
Interactive visual analysis of multidimensional geometric data
(Václav Skala - UNION Agency, 2016) Milman, Igal; Pilyugin, Victor V.; Skala, Václav
One of the most important tasks in modern world is to find solutions to problems of processing and analyzing multidimensional data. In this paper we present an approach for cluster analysis of multidimensional geometric data. Some definitions and extensions of classical cluster analysis problem is given. Our approach is based on the visualization method. Suggested approach allows us to analyze multidimensional data and distances in multidimensional Euclidean space using three-dimensional spatial scenes and shows an easy way for cluster analysis and anomaly discovery. An example of solving the problem of analysis of financial multidimensional data of credit organizations is also presented.
Multiphase action representation for online classification of motion capture data
(Václav Skala - UNION Agency, 2016) Salamah, Samer; Brunnett, Guido; Skala, Václav
In this paper we introduce a novel, simple, and efficient method for human action recognition based on a multiphase representation of human motion. An action is considered as a finite state machine where each state represents a primitive motion called motion phase, which is simply a sequence of poses with predefined common features. Spatial-temporal and postural features introduced in previous work are redefined by using only 3D joint positions for features extraction and are extended by involving the relative movement of the body end-effectors as new features. We developed a framework for modelling a given motion in the proposed motion model, whereupon we used this framework to create a model database of 25 different actions. Using this database we conducted a number of experiments on data obtained from several sources as well as on distorted data. The results showed that the presented method has high accuracy and efficiency. Additionally, it can work offline and online in real time, and can be easily adapted to work on 2D data.

Browse

Recent Submissions