Electronic Letters on Computer Vision and Image Analysis (ELCVIA - Universitat Autònoma de Barcelona)
Not a member yet
343 research outputs found
Sort by
Video Summarization by an Innovative Method in Shot Detection
Full text linkThe aim of a video summarization system is to provide a set of key frames which contain the most important parts of video. This method results in efficient storage, quick browsing, and retrieval of video collection. In this paper, we propose a new summarization system which firstly divides the video into meaningful shots using an innovative and fast method, and then we sample the video frames of each shot. This results in 97% reduction in under-process video frames. Then, using various characteristics of sampled frames such as color histogram, correlation and moment of inertia, we propose an adaptive aggregation function for combination of these characteristics (differences) and extraction of key frames. The proposed system is evaluated using 250 manual key frames constructed by human operators from 50 downloaded videos. The obtained results show that the proposed system provides better results compared to 6 different traditional methods
Tracheal Structure Characterization using Geometric and Appearance Models for Efficient Assessment of Stenosis in Videobronchoscopy
Full text linkRecent advances in endoscopic devices have increased their use for minimal invasive diagnostic and intervention procedures. Among all endoscopic modalities, bronchoscopy is one of the most frequent with around 261 millions of procedures per year. Although the use of bronchoscopy is spread among clinical facilities it presents some drawbacks, being the visual inspection for the assessment of anatomical measurements the most prevalent of them. In particular, inaccuracies in the estimation of the degree of stenosis (the percentage of obstructed airway) decreases its diagnostic yield and might lead to erroneous treatments. An objective computation of tracheal stenosis in bronchoscopy videos would constitute a breakthrough for this non-invasive technique and a reduction in treatment cost.This thesis settles the first steps towards on-line reliable extraction of anatomical information from videobronchoscopy for computation of objective measures. In particular, we focus on the computation of the degree of stenosis, which is obtained by comparing the area delimited by a healthy tracheal ring and the stenosed lumen. In this sense, we have to consider that reliable extraction of airway structures in interventional videobronchoscopy is a challenging task. This is mainly due to the large variety of acquisition conditions (positions and illumination), devices (different digitalizations) and in videosacquired at the operating room the unpredicted presence of surgical devices (such as probe ends). This thesis contributes to on-line stenosis assessment in several ways. We propose a parametric strategy for the extraction of lumen and tracheal rings regions based on the characterization of their geometry and appearance that guide a deformable model. The geometric and appearance characterization is based on a physical model describing the way bronchoscopy images are obtained and includes local and global descriptions. In order to ensure a systematic applicability we present a statistical framework to select the optimal parameters of our method. Experiments perform on the first public annotated database, show that the performance of our method is comparable to the one provided by clinicians and its computation time allows for a on-line implementation in the operating room
Automatic Segmentation of Optic Disc in Eye Fundus Images: A Survey
Full text linkOptic disc detection and segmentation is one of the key elements for automatic retinal disease screening systems. The aim of this survey paper is to review, categorize and compare the optic disc detection algorithms and methodologies, giving a description of each of them, highlighting their key points and performance measures. Accordingly, this survey firstly overviews the anatomy of the eye fundus showing its main structural components along with their properties and functions. Consequently, the survey reviews the image enhancement techniques and also categorizes the image segmentation methodologies for the optic disc which include property-based methods, methods based on convergence of blood vessels, and model-based methods. The performance of segmentation algorithms is evaluated using a number of publicly available databases of retinal images via evaluation metrics which include accuracy and true positive rate (i.e. sensitivity). The survey, at the end, describes the different abnormalities occurring within the optic disc region
A Survey on Human Emotion Recognition Approaches, Databases and Applications
Full text linkThis paper presents the various emotion classification and recognition systems which implement methods aiming at improving Human Machine Interaction. The modalities and approaches used for affect detection vary and contribute to accuracy and efficacy in detecting emotions of human beings. This paper discovers them in a comparison and descriptive manner. Various applications that use the methodologies in different contexts to address the challenges in real time are discussed. This survey also describes the databases that can be used as standard data sets in the process of emotion identification. Thus an integrated discussion of methods, databases used and applications pertaining to the emerging field of Affective Computing (AC) is done and surveyed.This paper presents the various emotion classification and recognition systems which implement methods aiming at improving Human Machine Interaction. The modalities and approaches used for affect detection vary and contribute to accuracy and efficacy in detecting emotions of human beings. This paper discovers them in a comparison and descriptive manner. Various applications that use the methodologies in different contexts to address the challenges in real time are discussed. This survey also describes the databases that can be used as standard data sets in the process of emotion identification. Thus an integrated discussion of methods, databases used and applications pertaining to the emerging field of Affective Computing (AC) is done and surveyed
Integrated Registration, Segmentation, and Interpolation for 3D/4D Sparse Data
Full text linkWe address the problem of object modelling from 3D and 4D sparse data acquired as different sequences which are misaligned with respect to each other. Such data may result from various imaging modalities and can therefore present very diverse spatial configurations and appearances. We focus on medical tomographic data, made up of sets of 2D slices having arbitrary positions and orientations, and which may have different gains and contrasts even within the same dataset. The analysis of such tomographic data is essential for establishing a diagnosis or planning surgery.Modelling from sparse and misaligned data requires solving the three inherently related problems of registration, segmentation, and interpolation. We propose a new method to integrate these stages in a level set framework. Registration is particularly challenging by the limited number of intersections present in a sparse dataset, and interpolation has to handle images that may have very different appearances. Hence, registration and interpolation exploit segmentation information, rather than pixel intensities, for increased robustness and accuracy. We achieve this by first introducing a new level set scheme based on the interpolation of the level set function by radial basis functions. This new scheme can inherently handle sparse data, and is more numerically stable and robust to noise than the classical level set. We also present a new registration algorithm based on the level set method, which is robust to local minima and can handle sparse data that have only a limited number of intersections. Then, we integrate these two methods into the same level set framework.The proposed method is validated quantitatively and subjectively on artificial data and MRI and CT scans. It is compared against a state-of-the-art, sequential method comprising traditional mutual information based registration, image interpolation, and 3D or 4D segmentation of the registered and interpolated volume. In our experiments, the proposed framework yields similar segmentation results to the sequential approach, but provides a more robust and accurate registration and interpolation. In particular, the registration is more robust to limited intersections in the data and to local minima. The interpolation is more satisfactory in cases of large gaps, due to the method taking into account the global shape of the object, and it recovers better topologies at the extremities of the shapes where the objects disappear from the image slices. As a result, the complete integrated framework provides more satisfactory shape reconstructions than the sequential approach
Modeling the environment with egocentric vision systems
Full text linkMore and more intelligent systems, such as robots or wearable systems, are present in our everyday life. This kind of systems interact with the environment so they need suitable models of their surrounding. Depending on the tasks that they have to perform, the information required in those models changes: fromhighly detailed 3D models for autonomous navigation systems, to semantic models including important information for the user. These models are created using the sensory data provided by the system (Fig. 1). Cameras are an important sensor included in most intelligent systems thanks to their small size, cheap prices and the great amount of information that they provide. This thesis studies and develops new methods to create models of the environment with different levels of precision and semantic information. There are two coomon key-points in the subsequent presented approaches:- The use of egocentric vision systems. All the vision systems and image sequences used in this thesis characterize for a first-person (egocentric) point of view.- The use of omnidirectional vision. This kind of vision systems provide much more information that conventional cameras thanks to their wide field of view.This thesis studies how computer vision can be used to create different models of the environment. To test our proposals, different cameras have been used, both in robotic and wearable platforms
A Novel Algorithm to Tackle Eyeglasses and Beard Issues in Facial IR Recognition
Full text linkFace recognition via thermal infrared (IR) images is a modern recognition method that has found so interesting for many researchers during last decade. This method which operates via thermal features and the situation of human face vessels has much more benefits than visual-based methods. In these images, the changes of environmental light, which is one of the most important problems of face recognition via visual images, are completely eliminated. The most important face recognition problem via thermal IR images is the existence of diffusion obstacles like glasses, which blocks an accurate extraction of the face vessels situation. Using the proposed algorithm, this problem has been completely removed. In this article face recognition is performed through face vessels. In fact, the proposed method solves the issues of face recognition (like glasses wearing) in the thermal infrared domain suggested by Pavlidis et al in [5]. For extraction of the face features, the situation of vessel branches is used. Also, by choosing appropriate classification, fake vessels and false branches are removed. On the other hand, the best feature is extracted by using Dynamic Time Wrapping (DTW) algorithm which is resistant to nonlinear changes. The simulation on UTK-IRIS gallery set shows the accurate recognition rate 95% on the images with glasses. Thus, the proposed method has improved the recognition rate about 10% on same gallery set compared to the best other methods
Bioinspired metaheuristics for image segmentation
Full text linkPhD thesis defended on 2nd December, 2013. In general, the purpose of Global Optimization (GO) is to find the global optimum of an objective function defined inside a search space, and it has applications in many areas of science, engineering, economics, among other, where mathematical modeling is used. GO algorithms are divided into two groups: deterministic and evolutionary. Since deterministic methods only provide a theoretical guarantee of locating local minimums of the objective function, they often face great difficulties in solving GO problems. On the other hand, evolutionary methods are usually faster in locating a global optimum than deterministic ones, because they operate on a population of candidate solutions, so they have a bigger likelihood of finding the global optimum, and even they have a better adaptation to black box formulations or complicated function’ forms. Even though during the last decade has had an important increasing in the area of metaheuristics applied to optimization, still is considered the searching of such methods as an open problem in research, due mainly to the fact that they still present difficulties, such as premature convergence and difficulty to overcome local optimums. Therefore, in this work it is proposed a bio inspired algorithm, who takes as inspiration the mechanism of allostasis. Allostasis is a biological term which explains how the modifications of specialized organ conditions inside the body allow achieving stability when an unbalance health condition is presented. If a body decompensation happens, according to the allostatic mechanisms, several body conditions compound by blood pressure, oxygen tension and others indexes are proved in order to get a stability state in health. By using the allostatic mechanisms as a metaphor, it is that we propose a metaheuristic algorithm, which we called Allostatic Optimization (AO). Such algorithm provides a searching procedure that is population-based, under which all the individuals, seen as body conditions, are defined in a multidimensional search space; aforementioned agents are either generated or modified by mean of several evolutionary operators who emulate the various operations used by the allostatic process, whereas an objective function evaluates the individual\u27s capacity (body condition) to reach a steady health state (good solution). AO is compared against DE, ABC and PSO and, different to them, the proposed algorithm favors the exploration process and eliminates some flaws related with premature convergence. By making such a comparison, it was found that in 57% of the functions the diversity maintained by AO helps the convergence of the algorithm, due to the fact that introduces operators that avoid particle concentration on some regions of the search space, favoring exploration. It was also found that maintaining a high diversity in the population does not guarantee the proper convergence of AO in all the benchmark functions, so a possible future work in this part of the investigation could be a more complete study of the relations among properties of functions, diversity and their relations with adequate convergence of the algorithm.AO was also used in image segmentation by using a mixture of functions; with the purpose of demonstrate the utility of the algorithm in a particular family of problems. One way to achieve segmentation is by mean of threshold selection, where each pixel that belongs to a determined class is labeled according to the selected threshold, giving as a result pixel groups that share visual characteristics in the image. In this work we use a method based of a mixture of Cauchy functions to approximate 1D histograms of gray level images, and it was found that AO improves the segmentation quality in about 14% when it was compared with Otsu’s method over known image benchmarks. Moreover, the metaheuristic algorithms DE, ABC and PSO were compared when they were applied to image segmentation by using a method that uses a mixture of Gaussian functions to approximate 1D histograms, because an analysis of such kind was not found in literature; the empirical results were that DE gives the best results in terms of convergence speed as well as quality of segmentation when compared against ground-truth images
Dynamic Obstacle Detection of Road Scenes using Equi-Height Mosaicking Image
Full text linkMany automobile companies and researchers have developed various safety systems to reduce fatalities by traffic accidents. In order to prevent traffic accidents by distracted driving, therefore, we present the vehicle and pedestrian detection using a novel image representation called equi-height mosaicking system. Furthermore, we additionally suggest the part-based side detection method using equi-height peripheral mosaicking image to detect approaching vehicles while drivin
Multimodal Stereo from Thermal Infrared and Visible Spectrum
Full text linkRecent advances in thermal infrared imaging (LWIR) has allowed its use in applications beyond of the military domain. Nowadays, this new family of sensors is included in different technical and scientific applications. They offer features that facilitate tasks, such as detection of pedestrians, hot spots, differences in temperature, among others, which can significantly improve the performance of a system where the persons are expected to play the principal role. For instance, video surveillance applications, monitoring, and pedestrian detection.During the dissertation the next question is stated: \textit{Could a couple of sensors measuring different bands of the electromagnetic spectrum, as the visible and thermal infrared, be used to extract depth information?} Although it is a complex question, we shows that a system of these characteristics is possible as well as their advantages, drawbacks, and potential opportunities.In this research an experimental study that compares different cost functions and matching approaches is performed, in order to build a multimodalstereovision system. Furthermore, the common problems in infrared/visible stereo, specially in the outdoor scenes are identified. Our framework summarizes the architecture of a generic stereo algorithm, at different levels: computational, functional, and structural, which can be extended toward high-level fusion (semantic) and high-order (prior).The proposed framework is intended to explore novel multimodal stereo matching approaches, going from sparse to dense representations (both disparity and depth maps). Moreover, context information is added in form of priors and assumptions. Finally, the dissertation shows a promissory way toward the integration of multiple sensors for recovering three-dimensional information