Electronic Letters on Computer Vision and Image Analysis (ELCVIA - Universitat Autònoma de Barcelona)
Not a member yet
343 research outputs found
Sort by
A PSO Application in Skull Prosthesis Modelling by Superellipse
Get PDFThis paper presents a method to create the geometric model of skull defects to be applied in anatomic prosthesis modelling. The approach is to generate an image that represents the missing information in the skull when bone`s defect is non-symmetric. We are proposing the use of superellipse concept to recover the parameters that represents the geometric shape of a skull bone curvature in tomography. If the superellipse is properly adjusted in each computed tomography slice, the arcs that represent the piece of missing bone can be modelled in 3D. The problem is that many similar ellipses can be created, and the best solution must be found. This research applies the Particle Swarm Optimization (PSO) algorithm in order to find the best solution for each tomographic slice. Once the solution found for each slice, the whole 3D missing information can be virtually rebuilt as an adjusted prosthesis model image
Comprehensive Analysis of High-Performance Computing Methods for Filtered Back-Projection
Get PDFThis paper provides an extensive analysis concerning runtime, accuracy and noise of High-Performance Computing (HPC) frameworks for Computed Tomography (CT) reconstruction tasks: "conventional" multi-core, multi threaded CPUs, the Compute Unified Device Architecture (CUDA) on GPUs, and the graphics pipeline of GPUs as facilitated by the DirectX or OpenGL programming interfaces, exploiting various built-in hardwired features like rasterization and texture filtering. We compare implementations of the Filtered Back-Projection (FBP) algorithm with fan-beam geometry on all these HPC frameworks. Specifically, an ACR-accredited phantom is reconstructed from the raw attenuation data acquired by a clinical CT scanner. Our analysis shows that a single GPU can run the FBP algorithm for reconstructing a 1024 x 1024 image considerably faster than a 64-core, multi-threaded CPU machine. Moreover, employing the graphics pipeline further increases performance as compared to CUDA, albeit with slightly lower accuracy due to "fast math" operations
Alzheimer\u27s disease early detection from sparse data using brain importance maps
Get PDFStatistical methods are increasingly used in the analysis of FDG-PET images for the early diagnosis of Alzheimer’s disease. We will demonstrate a method to extract information about the location of metabolic changes induced by Alzheimer’s disease based on a machine learning approach that directly relies features and brain areas to search for regions of interest (ROIs). This approach has the advantage over voxel-wise statistics to consider also the interactions between the features/voxels. We produce “maps” to visualize the most informative regions of the brain and compare the maps created by our approach with voxel-wise statistics. In classification experiments, using the extracted maps, we achieved classification rates of up to 95.5%
Perceptual Color Image Smoothing via a New Region-Based PDE Scheme
Get PDFIn this paper, we present a new color image regularization method using a rotating smoothing filter. This approach combines a method of pixel classification, which roughly determines if a pixel belongs to a homogenous region or an edge with an anisotropic perceptual edge detector capable of computing two precise diffusion directions. These directions are then used by an anisotropic diffusion scheme. Anisotropic diffusion is accurately controlled near edges and corners, while isotropic diffusion is applied to smooth regions either homogeneous or corrupted by noise. A comparison of our approach with other regularization methods applied on real images demonstrate that our model is able to efficiently restore images and control the diffusion, preserving well edges and corners
3D Segmentation for Multi-Organs in CT Images
Get PDFThe study addresses the challenging problem of automatic segmentation of the human anatomy needed for radiation dose calculations.Three-dimensional extensions of two well-known state-of-the art segmentation techniques are proposed and tested for usefulness on a set of clinical CT images.The new techniques are 3D Statistical Region Merging (3D-SRM) and 3D Efficient Graph-based Segmentation (3D-EGS). Segmentations of eight representative tissues (lungs, stomach, liver, heart, kidneys, spleen, bones and the spinal cord)were tested for accuracy using the Dice index, the Hausdorff distance and the index. The 3D-SRM outperformed 3D-EGS producing the average(across the 8 tissues) Dice index, the Hausdorff distance, and the of , ~mm and , respectively
A Colour Iris Recognition System Employing Multiple Classifier Techniques
Get PDFThe randomness of iris texture has allowed researchers to develop biometric systems with almost flawless accuracies. However, a common drawback of the majority of existing iris recognition systems is the constrained environment in which the user is enroled and recognized. The iris recognition systems typically require a high quality iris image captured under near infrared illumination. A desirable property of an iris recognition system is to be able to operate on colour images, whilst maintaining a high accuracy. In the present work we propose an iris recognition methodology which is designed to cope with noisy colour iris images. There are two main contributions of this paper: first, we adapt standard iris features proposed in literature for near infrared images by applying a feature selection method on features extracted from various colour channels; second, we introduce a Multiple Classifier System architecture to enhance the recognition accuracy of the biometric system. With a feature size of only 360 real valued components, the proposed iris recognition system performs with a high accuracy on UBIRISv1 dataset, in both identification and verfication scenarios
A novel framework for retrieval and interactive visualization of multimodal data
Get PDFWith the abundance of multimedia in web databases and the increasing user need for content of many modalities, such as images, sounds, etc., new methods for retrieval and visualization of multimodal media are required. In this paper, novel techniques for retrieval and visualization of multimodal data, i.e. documents consisting of many modalities, are proposed. A novel cross-modal retrieval framework is presented, in which the results of several unimodal retrieval systems are fused into a single multimodal list by the introduction of a cross-modal distance. For the presentation of the retrieved results, a multimodal visualization framework is also proposed, which extends existing unimodal similarity-based visualization methods for multimodal data. The similarity measure between two multimodal objects is defined as the weighted sum of unimodal similarities, with the weights determined via an interactive user feedback scheme. Experimental results show that the cross-modal framework outperforms unimodal and other multimodal approaches while the visualization framework enhances existing visualization methods by efficiently exploiting multimodality and user feedback
A New Weighted Region-based Hough Transform Algorithm for Robust Line Detection in Poor Quality Images of 2D Lattices of Rectangular Objects
Get PDFIn this work we present a novel kernel-based Hough Transform method for robust line detection in poor quality images of 2D lattices of rectangular objects. First, during a preprocessing stage, the connected regions of the image are determined. Then, a rectangularity score is computed for each region in order to filter out non-rectangular regions. Finally, the proposed method uses a kernel to specify each region’s contribution to the accumulator array based on the following shape descriptors: a) its rectangularity, b) the orientation of the major side of its minimum area bounding rectangle (MBR), and c) the MBR’s geometrical center. The proposed kernel is designed as the product of Gaussians having as footstep shape in Hough space that of a sinusoidal ribbon. Experimental and theoretical analysis on the uncertainties associated with the geometrical center as well as the polar parameters of the MBR’s major axis line equation allows for automatic selection of the parameters used to specify the shape of the kernel’s footstep (e.g. length and width of the ribbon) on the accumulator array. Comparisons performed on images of building facades taken under impaired visual conditions or with low accuracy sensors (e.g. thermal images) between the proposed method and other Hough Transform algorithms, show an improved accuracy of our method in detecting lines and/or linear formations. Finally, the robustness of the proposed method is shown in two other application domains those of, façade image rectification and skew detection and correction in rotated scanned documents
Generation and Rendering of Interactive Ground Vegetation for Real-Time Testing and Validation of Computer Vision Algorithms
Get PDFIn recent years, the realistic rendering of natural outdoor sceneries has become more and more important in many application areas. Regarding the development process of new algorithms for computer vision applications, testing and evaluation in real outdoor environments is time-consuming or hardly applicable in many cases. As a result,artificial testing environments are used, which differ from real-world environments, especially regarding realistically reacting ground vegetation. Thus, developers try to simulate natural environments in real-time virtual reality applications, which are commonly known as Virtual Testbeds. Since the first basic usage of Virtual Testbeds several years ago, the image quality of recent virtual environments has almost reached a level close to photorealism even in real-time due to new rendering approaches and increasing processing power of current graphics hardware. Thus, Virtual Testbeds can recently be applied in application areas like computer vision, that strongly rely on realistic scene representations. In this article, we introduce a novel ground vegetation rendering approach that is capable of generating large sceneries with realistic appearance and excellent performance. Our approach features wind animation, as well as object-to-grass interaction and delivers realistically appearing grass and shrubs at all distances and from all viewing angles. This greatly improves immersion, as well as acceptance, especially in virtual training applications. Nevertheless, the rendered results also fulfill important requirements for the computer vision aspect, like plausible geometry representation of the vegetation as well as its consistence during the entire simulation. Feature detection and matching algorithms are applied to our approach in localization scenarios of mobile robots in natural outdoor environments. We will show how the quality of feature matching results is influenced by violating the static scene constraint in the Virtual Testbed, as observed in highly unstructured, real-world outdoor scenes with wind and object-to-vegetation interaction
Detection and Classification of Multiple Objects using an RGB-D Sensor and Linear Spatial Pyramid Matching
Get PDFThis paper presents a complete system for multiple object detection and classification in a 3D scene using an RGB-D sensor such as the Microsoft Kinect sensor. Successful multiple object detection and classification are crucial features in many 3D computer vision applications. The main goal is making machines see and understand objects like humans do. To this goal, the new RGB-D sensors can be utilized since they provide real-time depth map which can be used along with the RGB images for our tasks. In our system we employ effective depth map processing techniques, along with edge detection, connected components detection and filtering approaches, in order to design a complete image processing algorithm for efficient object detection of multiple individual objects in a single scene, even in complex scenes with many objects. Besides, we apply the Linear Spatial Pyramid Matching (LSPM) [1] method proposed by Jianchao Yang et al for the efficient classification of the detected objects. Experimental results are presented for both detection and classification, showing the efficiency of the proposed design.