WSCG '2015: Short Papers Proceedings

Permanent URI for this collection

http://hdl.handle.net/11025/26399

Browse

Showing 1 - 20 out of 21 results

Inpainted image quality assessment based on machine learning
(Václav Skala - UNION Agency, 2015) Voronin, V.; Marchuk, V.; Semenishchev, E.; Maslennikov, S.; Svirin, I.; Skala, Václav
In many cases inpainting methods introduce a blur in sharp transitions in image and image contours in the recovery of large areas with missing pixels and often fail to recover curvy boundary edges. Quantitative metrics of inpainting results currently do not exist and researchers use human comparisons to evaluate their methodologies and techniques. Most objective quality assessment methods rely on a reference image, which is often not available in inpainting applications. This paper focuses on a machine learning approach for noreference visual quality assessment for image inpainting. Our method is based on observation that Local Binary Patterns well describe local structural information of the image. We use a support vector regression learned on human observer images to predict the perceived quality of inpainted images. We demonstrate how our predicted quality value correlates with qualitative opinion in a human observer study.
Robust hand gesture recognition from 3D data
(Václav Skala - UNION Agency, 2015) Kurmi, Vinod K; Jain, Garima; Venkatesh, KS; Skala, Václav
In this paper, we use the output of a 3D sensor (ex. Kinect from Microsoft) to capture depth images of humans making a set of predefined hand gestures in various body poses. Conventional approaches using Kinect data have been constrained by the limitation of the human detector middleware that requires close conformity to a standard near erect, legs apart, hands apart pose for the subject. Our approach also permits clutter and possible motion in the scene background, and to a limited extent, in the foreground as well. We make an important point in this work to emphasize that the recognition performance is considerably improved by a choice of hand gestures that accommodate the sensor’s specific limitations. These sensor limitations include low resolution in x and y as well as z. Hand gestures have been chosen(designed) for easy detection by seeking to detect a fingers apart, fingertip constellation with minimum computation. without, however compromising on issues of utility or ergonomy. It is shown that these gestures can be recognised in real time irrespective of visible band illumination levels, background motion, foreground clutter, user body pose, gesturing speeds and user distance. The last is of course limited by the sensor’s own range limitations. Our main contributions are the selection and design of gestures suitable for limited range, limited resolution 3D sensors and the novel method of depth slicing used to extract hand features from the background. This obviates the need for preliminary human detection and enables easy detection and highly reliable and fast (30 fps) gesture classification.
Analysis of 3D mesh correspondences concerning foldovers
(Václav Skala - UNION Agency, 2015) Merz, Johannes; Getto, Roman; Landesberger, Tatiana von; Fellner, Dieter W.; Skala, Václav
Foldovers (i.e., folding of triangles in a 3D mesh) are artifacts that cause problems for morphing. Mesh morphing uses vertex correspondences among the source and the target mesh to define the morphing path. Although there exist techniques for making a foldover-free mesh morphing, identification and correction of foldovers in existing correspondences is still an unsolved issue. This paper proposes a new technique for the identification and resolution of foldovers for mesh morphing using predefined 3D mesh correspondences. The technique is evaluated on several different meshes with given correspondences. The mesh examples comprise both real medical data and synthetically deformed meshes. We also present various possible usage scenarios of the new algorithm, showing its benefit for the analysis and comparison of mesh correspondences with respect to foldover problems.
A geodesic based approach for an accurate and invariant 3D surfaces representation
(Václav Skala - UNION Agency, 2015) Jribi, Majdi; Ghorbel, Faouzi; Skala, Václav
In this paper, we propose a novel 3D invariant surface representation under the 3D motion group M(3). It is obtained by combining two main representations: the three-polar representation and the one defined by the radial line curves from a starting point. The retained invariant points correspond to the geometrical locations of the intersection between the two last representations. The approximation of the novel surfaces description method on the 3D discrete meshes is studied. Its accuracy for the 3D faces description and retrieval is evaluated in the mean of the Hausdorff shape distance.
News Patterns: how press interacts with social networks
(Václav Skala - UNION Agency, 2015) Boettcher, Guilherme Bertini; Comba, João; Dal Sasso Freitas, Carla; Skala, Václav
Social media has played a big part in the adaptation process for newspapers and magazines, but innovating while going through a recession has led to a hasty evolution and automated processes for very different media. While existing socialmedia studies and state of the art visual solutions are available for analyzing social media content and users’ behaviors, no other method is optimized for finding patterns from a popularity standpoint in the specialized realm of news channels. In this paper, we propose the usage of a combination of different visualization techniques that co-relate the profile’s and its reading community activities with the resulting popularity. For the period of three months, we gathered Twitter posts, the number of followers and trending topics from worldwide press profiles. We used this dataset as the seed for our bar charts, tag clouds and bubble charts to allow for multiple source comparison, so that not only the user is able to understand their own community but also the success and pitfalls faced by the competition in the same medium. We validate our analysis by interviewing a group of journalists from different established newspapers. Through interacting with our system, it was possible to reveal hidden patterns in the massive dataset of messages and comments worldwide enabling the user to have unique insight into their community’s behaviors and preferences.
Perspective correction of panoramic images created by parallel motion stitching
(Václav Skala - UNION Agency, 2015) Gonçalves, João; Ribeiro, David; Soares, Filipe; Skala, Václav
This paper deals with the problem of correcting the perspective distortion in panoramic images created by parallel motion stitching. The distortion is revealed by lines that appear to converge at the infinity, but are actually parallel. A camera cart shoots from multi-viewpoints aiming a parallel motion to the scene that is photographed. The perspective effect arises on panoramas while stitching several images taken from the camera, slightly panning in both directions between shots along the motion path. In this paper, we propose a solution to handle different camera translation motions and be able to stitch together images with a high-level of similarity, also having repetition patterns along a vast continuity of elements belonging to the scene. The experimental tests were performed with real data obtained from supermarket shelves, with the goal of maintaining the correct amount of product items on the resulting panorama. After applying the perspective correction in the input images, to reduce cumulative registration errors during stitching, it is possible to extract more information about the similarity between consecutive images so that matching mistakes are minimized.
VideoMR: a map and reduce framework for real-time video processing
(Václav Skala - UNION Agency, 2015) Meier, Benjamin-Heinz; Trapp, Matthias; Döllner, Jürgen; Skala, Václav
This paper presents VideoMR: a novel map and reduce framework for real-time video processing on graphic processing units (GPUs). Using the advantages of implicit parallelism and bounded memory allocation, our approach enables developers to focus on implementing video operations without taking care of GPU memory handling or the details of code parallelization. Therefore, a new concept for map and reduce is introduced, redefining both operations to fit to the specific requirements of video processing. A prototypic implementation using OpenGL facilitates various operating platforms, including mobile development, and will be widely interoperable with other state-of-the-art video processing frameworks.
Surfaces for point clouds using non-uniform grids on the GPU
(Václav Skala - UNION Agency, 2015) Schiffner, Daniel; Stockhausen, Claudia; Ritter, Marcel; Skala, Václav
Clustering data is a standard tool to reduce large data sets, such as scans from a LiDAR, enabling real-time rendering. Starting from a uniform grid, we redistribute points from and to neighboring cells. This redistribution is based on the properties of the uniform grid and thus the grid becomes implicitly curvilinear, which produces better matching representatives. Combining these with a polygonal surface reconstruction enables us to create interactive renderings of dense surface scans. Opposed to existing methods, our approach is running solely on the GPU and is able to use arbitrary data fields to influence the curvilinear grid. The surfaces are also generated on the GPU to avoid unnecessary data storage. For evaluation, different data sets stemming from engineering and scanning applications were used and have been compared against typical CPU based reconstruction methods in terms of performance and quality. The proposed method turned out to reach interactivity for large sized point clouds, while being able to adapt to the point clouds geometry, especially when using non-uniform sampled data.
Edge-aware normal estimation by rotated bilateral sampling
(Václav Skala - UNION Agency, 2015) Kovacs, Viktor; Tevesz, Gabor; Skala, Václav
In this paper we deal with edge preserving surface normal estimation and crease edge detection in discretized range images. Such range images consist of few discrete quantization levels due to the data acquisition method (short base distance stereo), or when the distance variation of the examined surface is low, compared to the disparity quantization levels. We propose a method for normal estimation and crease edge detection using iso-range curves and rotated bilateral filter based sampling. Iso-range curves are used to extract sparse, but reliable range image points. Samples are first selected by a rotated weight matrix and a plane is fitted on such samples. Simple statistics are gathered during the rotation of the weight matrix, in order to find the best fitting plane and extract crease edge measure. Such information may be used for further range image processing: segmentation, mapping, localization, object detection, recognition etc. Results are shown for both synthetic and real range images. It was shown that applying the proposed method resulted in more accurate normal estimations, crease edges were not smoothed and crease edges were successfully detected.
Interactive tool and database for optic disc and cup segmentation of stereo and monocular retinal fundus images
(Václav Skala - UNION Agency, 2015) Fumero, Francisco; Sigut, Jose; Alayón, S.; González-Hernández, M.; González de la Rosa, M.; Skala, Václav
Glaucoma is one of the leading causes of irreversible blindness in the world. Early detection is essential to delay its evolution and avoid vision loss. For this purpose, retinal fundus images can be used to assess the cup-to-disc ratio, the main indicator of glaucoma. Several automatic methods have been developed to compute this indicator, but the lack of ground truth of the optic disc and cup is an obstacle to evaluate and compare their results. In order to support clinicians to perform this task, an interactive tool for the segmentation of the disc and cup on stereo and monocular retinal fundus images has been developed. By using this tool, we have also built a new database of 159 stereo fundus images with two ground truth of disc and cup. The application and the database are both publicly available online. This work can serve as a learning environment for clinicians, as well as to evaluate the results of automatic segmentation algorithms.
Over- and under-segmentation evaluation based on the segmentation covering measure
(Václav Skala - UNION Agency, 2015) Sigut, Jose; Fumero, Francisco; Nuñez, Omar; Skala, Václav
Very few measures intended for evaluating the quality of image segmentations account separately for over- and under-segmentation. This distinction is highly desirable in practice because in many applications undersegmentation is considered as a much serious issue than over-segmentation. In this paper, a new approach to this problem is presented as a decomposition of the Segmentation Covering measure into two contributions, one due to over-segmentation and the other one to under-segmentation. Our proposal has been tested on the output of state-of-the-art segmentation algorithms using the Berkeley image database. The results obtained are comparable to those provided by similar evaluation methods allowing a clear separation between over- and undersegmentation effects.
Hyperspectral image xlassification using a general NFLE transformation with kernelization and fuzzification
(Václav Skala - UNION Agency, 2015) Chen, Ying-Nong; Wang, Yu-Chen; Han, Chin-Chuan; Fan, Kuo-Chin; Skala, Václav
Nearest feature line (NFL) embedding (NFLE) is an eigenspace transformation algorithm based on the NFL strategy. Based on this strategy, the NFLE algorithm generates a low dimensional space in which the local structures of samples in the original high dimensional space are preserved. Though NFLE has successfully demonstrated its discriminative capability, the non-linear manifold structure cannot be structured more efficiently by linear scatters using the linear NFLE method. To address this, a general NFLE transformation, called fuzzy/kernel NFLE, is proposed for feature extraction in which kernelization and fuzzification are simultaneously considered. In the proposed scheme, samples are projected into a kernel space and assigned larger weights based on that of their neighbors according to their neighbors. In that way, not only is the non-linear manifold structure preserved, but also are the discriminative powers of classifiers increased. The proposed method is compared with various state-of-the-art methods to evaluate the performance by several benchmark data sets. From the experimental results, the proposed FKNFLE outperformed the other, more conventional, methods.
Introducing aesthetics to software visualization
(Václav Skala - UNION Agency, 2015) Baum, David; Skala, Václav
In software visualization, but also in information visualization in general, there is a great need for evaluation of visualization metaphors. To reduce the amount of empirical studies a computational approach has been applied successfully, e.g., to graph visualization. It is based on measurable aesthetic heuristics that are used to estimate the human perception and the processing of visualizations. This paper lays a foundation for adopting this approach to any field of information visualization by providing a method, the repertory grid technique, to identify aesthetics that are measurable, metaphor-specific, and relevant to the user in a structured and repeatable way. We identified 25 unique aesthetics and revealed that the visual appearance of the investigated visualizations is mainly influenced by the package structure whereby methods are underrepresented. These findings were used to improve existing visualizations.
3D reconstruction of outdoor scenes using structure from motion and depth data
(Václav Skala - UNION Agency, 2015) Fujimoto, Keisuke; Watanabe, Takashi; Skala, Václav
Recently, low-cost and small RGB-D sensors appear massively at the entertainment market. These sensors can acquire colored 3D models using color images and depth data. However, a limitation of the RGB-D sensor is that sunlight interferes with the pattern projecting LED. The sensor is most suitable only for indoor scenes. Some RGBD sensors are available in outdoor scenes. However, the measurement range is limited because the light of LED spreads in all directions. In this research, we developed a novel measurement method for RGB-D sensors, which can measure shapes in outdoor scenes. This method uses several measurement data from multiple viewpoints, and estimates the shape and the sensor poses using Structure from Motion (SfM). However, a conventional imagebased SfM cannot determine a correct scale. To determine the correct scale, our method uses the depth information that is obtained from partially acquired area which is near to the viewpoints. Then, our method optimizes the shape and the poses by a modified bundle adjustment with the depth information. It minimizes the reprojection error of the features in the acquired images and the depth error between the estimated model and the measurement depth. At last, our method generates dense point cloud using a multi-view stereo algorithm. Using both the acquired images and depth data, our method reconstructs the shape which locates out of measurement range in outdoor environment. In our experiment, we show that our method can measure the range up to 20 meters away by measuring from several viewpoints in the range of 5 meters using a RGB-D sensor in outdoor scenes.
Visual encoding of automatic identification system (AIS) data for radar systems
(Václav Skala - UNION Agency, 2015) Last, Philipp; Hering-Bertram, Martin; Jung, Thomas; Linsen, Lars; Skala, Václav
The Automatic Identification System (AIS) is a maritime system mostly used for automatically exchanging tracking and other relevant information between vessels. It supports decision making of nautical personnel such as master mariners. AIS data are multivariate including many aspects for identification and localization of ships and for navigation. However, during navigation not all AIS data are made visually available to the nautical personnel. In this paper, we propose a glyph-based visualization consistent with currently used encodings for intuitively and effectively encoding further so far missing AIS data attributes on radar screens. Proposed extensions aim at increasing maritime safety by helping mariners to assess traffic situations. We applied our visualization methods to real-world data recorded at the German North Sea coast and evaluated them with the help of an expert group.
Anomaly detection using spectral mismatch between anomaly pattern and its neighborhood
(Václav Skala - UNION Agency, 2015) Denisova, A.Yu.; Myasnikov, Vladislav; Skala, Václav
In this paper we present a novel algorithm for anomaly detection in multichannel images. Proposed algorithm uses spectral mismatch criterion to describe anomalous properties of small image regions. The idea behind the criterion is that the brightness of the anomalous region can't be represented as a function of pixels comprising that region. In our paper, we consider a local pattern of anomaly and its neighborhood, and we use a linear function to approximate the anomaly at each image position. In contrast to existing global and local RXD algorithms our approach allows more adaptive and noise resistant detection of anomalies. Experimental results are presented for hyperspectral remote sensing images.
Efficient linear local features of digital signals and images: computational and qualitative properties
(Václav Skala - UNION Agency, 2015) Myasnikov, Vladislav; Skala, Václav
The paper presents the analysis of efficiency of two original approaches to the construction of the sets of linear local features (LLF), which are used for digital signal and image processing. The first approach is based on generating of LLF set, which consists of separately constructed efficient LLFs, each of which has its own algorithm for feature calculation. The second approach assumes the construction of an efficient LLF set, which has a single algorithm for joint simultaneous computation of all features. The analysis is carried out by several indicators that characterize the computational and qualitative properties of the constructed LLFs.
Applying filters to repeating motion based trajectories for video classification
(Václav Skala - UNION Agency, 2015) Ayyildiz, Kahraman; Conrad, Stefan; Skala, Václav
The presented video classification system is based on the trajectory of repeating motion in video scenes. Further on this trajectory has a certain direction and velocity at each time frame. As the position, direction and velocity of the motion trajectory evolve in time, we consider these as motion functions. Later on we transform these functions by FFT and receive frequency domains, which then represent the frequencies of repeating motion. Moreover these frequencies serve as features during classification phase. Our current work focuses on filtering the functions based on the motion’s trajectory in order to reduce noise and emphasize significant parts.
3D avatar for automatic synthesis of signs for the sign languages
(Václav Skala - UNION Agency, 2015) Gonçalves, Diego Addan; Todt, Eduardo; Sanchez Garcia, Laura; Skala, Václav
This paper discusses a synthesis system that generates, from a XML input representing gesture descriptors, a vector of configuration parameters that are executed by a 3D Avatar for use in the animation of Sign Languages. The development of virtual agents able to reproduce gestures of sign languages is very important to the deaf community, since in general they also have difficulties to read conventional texts. In this research project, a consistent combination of 3D editor Blender, CMarkup parser and graphics engine Irrlicht was used to develop a novel approach to sign synthesis, based on a recent XML model that describes hand gestures using shape, location, movement and orientation descriptors. The described experiments validate the proposed implementation model, which constitutes a promising alternative in the area of synthesis of signals for computational applications of Sign Languages.
A framework for robust object multi-detection with a vote aggregation and a cascade filtering
(Václav Skala - UNION Agency, 2015) Kurzejamski, Grzegorz; Zawistowski, Jacek; Sarwas, Grzegorz; Skala, Václav
This paper presents a framework designed for the multi-object detection purposes and adjusted for the application of product search on the market shelves. The framework uses a single feedback loop and a pattern resizing mechanism to demonstrate the top effectiveness of the state-of-the-art local features. A high detection rate with a low false detection chance can be achieved with use of only one pattern per object and no manual parameters adjustments. The method incorporates well known local features and a basic matching process to create a reliable voting space. Further steps comprise of metric transformations, graphical vote space representation, two-phase vote aggregation process and a cascade of verifying filters.

Browse

Recent Submissions