[show abstract][hide abstract] ABSTRACT: Object recognition is an essential task in content-based image retrieval and classification. This paper deals with object recognition in WIKImage data, a collection of publicly available annotated Wikipedia images. WIKImage comprises a set of 14 binary classification problems with significant class imbalance. Our approach is based on using the local invariant image features and we have compared 3 standard and widely used feature types: SIFT, SURF and ORB. We have examined how the choice of representation affects the k-nearest neighbor data topology and have shown that some feature types might be more appropriate than others for this particular problem. In order to assess the difficulty of the data, we have evaluated 7 different k-nearest neighbor classification methods and shown that the recently proposed hubness-aware classifiers might be used to either increase the accuracy of prediction, or the macro-averaged F-score. However, our results indicate that further improvements are possible and that including the textual feature information might prove beneficial for system performance.
[show abstract][hide abstract] ABSTRACT: This paper presents a new automatic image annotation algorithm. First, we introduce a new similarity measure between images: compactness. This uses low level visual descriptors for determining the similarity between two images. Compactness indicates how close test image features lie to training image feature cluster centers. The measure provides the core for a k-nearest neighbor type image annotation method. Afterwards, a formalism for defining different transfer techniques is devised and several label transfer techniques are provided. The method as whole is evaluated on four image annotation benchmarks. The results on these sets validate the accuracy of the approach, which outperforms many state-of-the-art annotation methods. The method presented here requires a simple training process, efficiently combines different feature types and performs better than complex learning algorithms, even in this incipient form. The main contributions of this work are: the usage of compactness as a similarity measure which enables efficient low level feature comparison and an annotation algorithm based on label transfer.
IEEE Transactions on Image Processing 08/2013; · 3.20 Impact Factor
[show abstract][hide abstract] ABSTRACT: PurposeNoninvasive diagnosis of liver fibrosis is a popular topic in the medical literature. Textural analysis on B-mode ultrasound
is viewed as a noninvasive tool for fibrosis staging. A liver tissue model is proposed and used to simulate ultrasound images.
MethodsOne hundred and twenty-five patients with chronic hepatitis C were included in this study. Patients were investigated using
B-mode ultrasound and liver biopsy (Metavir scoring). A texture analysis tool consisting of 12 algorithms and a logistic regression
classifier was implemented and validated. Tissue model parameters were varied and ultrasound images were generated.
ResultsTexture analysis can discriminate between stages F0 and F4 using actual patient data (accuracy 69.5%) and synthetic images
(accuracy 76.6%). A human expert is less sensitive than texture analysis in discriminating subtle changes in ultrasound images.
High fibrosis detection accuracies are correlated with larger differences in portal space density (r
2=0.5). Accuracies measured when we varied only the fibrosis stage and kept the rest of the tissue parameters constant showed
high detection rates only in a narrow parameter interval.
ConclusionThe texture analysis system shows limited performance in staging fibrosis and it cannot be used for accurate monitoring of
fibrosis evolution over time.
KeywordsTissue model–Fibrosis staging–Noninvasive diagnosis–Texture analysis
Journal of Medical Ultrasonics 04/2012; 38(3):105-117. · 0.64 Impact Factor
[show abstract][hide abstract] ABSTRACT: Texture analysis is viewed as a method to enhance the diagnosis power of classical B-mode ultrasound image. The present paper aims to evaluate and eliminate the dependence between the human expert and the performance of such a texture analysis system in predicting the cirrhosis in chronic hepatitis C patients. 125 consecutive chronic hepatitis C patients were included in this study. Ultrasound images were acquired from each patient and four human experts established regions of interest. Textural analysis tool was evaluated. The performance of this approach depends highly on the human expert that establishes the regions of interest (P < 0.05). The novel algorithm that automatically establishes regions of interest can be compared with a trained radiologist. In classical form met in the literature, the noninvasive diagnosis through texture analysis has limited utility in clinical practice. The automatic ROI establishment tool is very useful in eliminating the expert-dependent variability.
Computational and Mathematical Methods in Medicine 01/2012; 2012:346713. · 0.79 Impact Factor
[show abstract][hide abstract] ABSTRACT: The suitability of stereo algorithms for intelligent vehicle applications is conditioned by their ability to compute dense accurate disparity maps in real time. In this paper, an original stereo reconstruction system that is designed for automotive applications is presented. The system is based on the semiglobal matching algorithm (SGM), which is widely known for its high quality and potential for real-time implementation. Several improvements that target the matching, disparity optimization, and disparity refinement steps are proposed. Pixel-level matching uses the census transform because of its invariance to intensity differences due to camera bias or gain that affects the images. The huge memory bandwidth requirements for the SGM disparity optimization step are reduced through a new integration strategy. At the subpixel level, accuracy is increased by devising a new methodology for generating dedicated subpixel interpolation functions. Using this methodology, two novel subpixel interpolation functions for the SGM algorithm are implemented and evaluated. The proposed algorithm has been implemented on a graphics processing unit in the Compute Unified Device Architecture (CUDA). The result is an increased speed and accuracy algorithm profiled for complex real-traffic scenarios. The proposed algorithm has been evaluated at a large scale, and evidence that was collected from both standard benchmarks and real-world images confirm the findings and show a significant improvement over existing solutions.
IEEE Transactions on Vehicular Technology - IEEE TRANS VEH TECHNOL. 01/2012; 61(3):1032-1042.
[show abstract][hide abstract] ABSTRACT: This paper presents an occupancy grid tracking system based on particles, and the use of this system for dynamic obstacle detection in driving environments. The particles will have a dual nature they will denote hypotheses, as in the particle filtering algorithms, but they will also be the building blocks of our modeled world. The particles have position and speed, and they can migrate in the grid from cell to cell depending on their motion model and motion parameters, but they will also be created and destroyed using a weighting-resampling mechanism specific to particle filter algorithms. An obstacle grid derived from processing a stereovision-generated elevation map is used as measurement information, and the measurement model takes into account the uncertainties of the stereo reconstruction. The dynamic occupancy grid is used for improving the quality of the stereovision-based reconstruction as oriented cuboids. The resulted system is a flexible, real-time tracking solution for dynamic unstructured driving environments, and a useful tool for extracting intermediate dynamic information that can considerably improve object detection and tracking.
[show abstract][hide abstract] ABSTRACT: Modeling the performance of large scale systems is the core idea of this paper. We focus on modeling the performance specific behavior of LarKC 1 - The Large Knowledge Collider a platform for large scale integrated reasoning and Web-search. A set of instrumentation and monitoring tools are employed to collect metrics related to execution time, resources, and specific platform measurements like running workflows and plug-ins. Our method performs machine learning on top of instrumented data and tries to find relations between input defined metrics and output metrics that describe the instrumentation observations of the LarKC platform, plug-ins or workflows. The proposed method is a combination of clustering and regression techniques.
[show abstract][hide abstract] ABSTRACT: This paper presents an approach for automatic recognition of interest objects from the low earth orbit that are visible in astronomical images. The main interest objects are satellites, but various objects, such as planes, stars, can be identified. The proposed technique starts with background estimation and removal. Then, potential objects are identified by labeling the background free image. Relevant features are computed for each individual object and used for classification, which is performed with a decision tree. Real and relevant data was used to evaluate the performance of this methodology and to determine its main strengths and weaknesses.
[show abstract][hide abstract] ABSTRACT: This paper presents a method for the detection and localization of stop-lines and other horizontal road markings as bicycle or pedestrian crossings encountered in road environments. The few existing stop-line detection approaches found in literature are based on monocular vision, and therefore cannot infer high accuracy information from the visual data. The proposed solution uses a hybrid approach that combines 2D detections in grayscale images with stereo-vision based 3D validations in order to increase the robustness and accuracy of the results. Model based reasoning is also used to eliminate false positive situations and to detect exactly the objects of interest (stop-lines, pedestrian or bicycle crossings). The detected horizontal road markings can be used as landmarks for a very accurate longitudinal localization of the ego vehicle within intersection scenarios compensating missing GPS data or its lack of accuracy. An important task in these scenarios is the accurate ego- vehicle localization relative to the intersection. Although GPS based digital maps, with static or even dynamic content have been the solution world-wide adopted for navigation both in urban and non-urban environments, the conventional GPS signal is affected by various factors that introduce significant positioning errors (between 1 and 30 meters), and thus a more accurate localization of the ego-vehicle on a digital map is necessary. This can be obtained by sensorial localization of some road landmarks relative to the ego car. In this paper a very accurate method for the longitudinal localization of the ego vehicle relative to an intersection is proposed. It is based on the detection and localization of some specific intersection landmarks: primarily stop-lines or wait- lines and secondarily other transversal road markings as pedestrian and bicycle crossings. Vision sensors are extensively used in the detection of the painted road markings since they provide the highest spatial resolution compared to other sensors as laser scanners or LIDARS. Approaches based on monocular vision can provide accurate 2D detection of the road markings but they cannot provide alone localization information. In (1) a stop-line detection algorithm is presented that uses the Canny edge detector to search each image for properly oriented pairs of edges, one a transition from dark road to white stop line paint and its pair, a transition from white paint back to road. In (2) painted road signs are segmented in the IPM image and are recognized using a neural network classifier applied on the extracted features. In (3) a complete solution for detection, measurement and classification of painted road objects that exhibit mainly vertical/longitudinal features as arrows, lane delimiters and zebras is presented. The road markings are segmented in the 2D grayscale image while the 3D information is inferred from the dense-stereo map provided by a stereo- vision sensor (4). The road markings are finally classified using a decision tree based classifier. The proposed solution for detecting stop-lines, wait-lines, pedestrian and bicycle crossings combines the detection of horizontal features in the 2D image followed by 3D validation/classification. The detection of the 2D image features is based on the Hough transform which is used to detect only horizontal lines from the edge map of a grayscale image. These horizontal lines can belong to different classes of objects, as barriers (5), road features (transversal road markings, curbs etc), or other unclassifiable horizontal structures. Therefore, a series of validations steps based on primary dense 3D data provided by a dense stereo sensor (4), road surface parameters provided by a lane detection module (7) and model based reasoning applied on the markings' 2D shape and 3D size are performed for an accurate detection and localization. The identified transversal road markings are reported as flat 3D cuboids which are further tracked using the odometry data of the ego car.
[show abstract][hide abstract] ABSTRACT: The hepatocellular carcinoma (HCC) is the most frequent malignant liver tumor. The golden standard for HCC diagnosis is the needle biopsy, but this is an invasive, dangerous method. We aim to develop computerized, non-invasive techniques for the automatic diagnosis of HCC, based on information obtained from ultrasound images. The texture is an important property of the internal organs tissue, able to provide subtle information about the pathology. We previously defined the textural model of HCC, consisting in the exhaustive set of the relevant textural features, appropriate for HCC characterization and in the specific values of these features. In this work, we analyze the role that the superior order Gray Level Cooccurrence Matrices (GLCM) and the associated parameters have in the improvement of HCC characterization and automatic diagnosis. We also determine the best spatial relation between the pixels that leads to the highest performances, for the third and fifth order GLCM.
34th International Conference on Telecommunications and Signal Processing (TSP 2011), Budapest, Hungary, Aug. 18-20, 2011; 01/2011
[show abstract][hide abstract] ABSTRACT: This paper presents an automatic annotation method for multimedia data. Different object and scene recognition methods are analyzed from the literature. The best components of current methods are used to design and implement an original solution. A novel approach for refining results based on scene specific object appearance frequencies is exposed which improves annotation performance. Experimental results indicate the performance of the proposed solution which successfully competes with currently best methods.
[show abstract][hide abstract] ABSTRACT: The quality of images in real time vision tends to be of extreme importance. Without high quality images processing of the outside world can introduce extreme errors. Image quality is dependent on many environmental factors, such as weather and lighting conditions (sun, rain, snow, fog, mist), entering and going out of tunnels, shadows, cars headlights. This paper presents a new approach for adapting the camera response with respect to the environment's lighting conditions. For this we model a digital camera's response as a function of its parameters. The obtained mathematical model is vital for adapting the cameras to the environment's lighting conditions. The most widely used camera parameters are the exposure time and amplification gain. By adjusting these parameters the image acquisition system can be less dependent on the environment's lighting conditions and can provide better quality images for further processing.
[show abstract][hide abstract] ABSTRACT: The hepatocellular carcinoma (HCC) is the most frequent malignant liver tumor. Nowadays, the only reliable method for the detection of HCC is the needle biopsy, but it is invasive, dangerous for the patient. We aim to develop a non-invasive method for the automatic diagnosis of HCC, based only on computerized techniques for ultrasound image analysis. Thus, we elaborated the imagistic textural model of HCC, consisting in the exhaustive set of the textural parameters, relevant for HCC characterization, and in their specific values for the HCC class. In this work, we study the effect of the classifier combination procedures on the improvement of the recognition performance, from speed and accuracy points of view. Various combination schemes are considered, and their influence on the accuracy parameters and on the learning curves is discussed. The hepatocellular carcinoma is also divided into subclasses, and the multiclass classification techniques are experimented for accuracy improvement.
[show abstract][hide abstract] ABSTRACT: We propose an environment representation technique by Temporal Analysis of the Occupancy Grid using a Dense Stereo-Vision System. The proposed method takes into account both the 3D information provided by the Occupancy Grid and the ego-car parameters. We use a method for computing the differences between the previous and current frames and compute an evidence space called Occupancy Grid Difference Map. Based on the difference map we created a reasoning component to generate an improved 2.5D model by representing the environment as a set of polylines with the associated static and dynamic features.
[show abstract][hide abstract] ABSTRACT: This paper proposes a two-level approach to the multiple objects tracking problem. One particle filter-based tracker will search the whole state space for new hypotheses, and when a hypothesis becomes strong enough, an individual object tracker will be activated, which will track it until the object is lost. The initialization tracker and the individual object trackers use two types of cues, a stereovision-based obstacle grid and a symmetry measure computed by grayscale image processing. The proposed solution is a simple and robust one, adaptable to different types of object models and extendable to different types of sensors.
[show abstract][hide abstract] ABSTRACT: This paper presents a new type of SLAM for indoor and outdoor environments. This method solves only the simplest localization problem, position tracking, and does not use the mapping information in localization. The proposed SLAM uses hybrid odometry for the localization. Mapping is based on LDPDs (local differential probability distances) and accumulated 3D dense frames.
[show abstract][hide abstract] ABSTRACT: This paper presents a lane estimation technique based on the particle filter framework, which fuses several image-based cues (edges, lane markings and curbs), and 3D cues extracted from stereovision. A partition sampling-like approach is used to decouple pitch estimation from the rest of the parameter set, allowing the use of a significantly lower number of particles, and initialization samples are used for faster handling of discontinuous roads. We also introduce a measure for detection quality, for result validation. The resulted solution has proven to be a reliable and fast lane detector for difficult scenarios.
[show abstract][hide abstract] ABSTRACT: This paper presents a new obstacle detection method that is achieved on the depth image. This obstacle detection method can be used both in clumsy office environments and in natural environments. In this paper we focus our attention on the obstacle detection in the case of pitch angle variation (weak calibration conditions) due to uneven-road surface. The proposed obstacle detection has, mainly, two characteristics. One is the time efficiency aspect in the labelling of detected obstacles, which is achieved on the binarized depth image. The contour pixels of the obstacles of the binarized depth image have 3D information. The other is flexibility aspect in road plane that is generated by the LDPD (local difference probability distance) road classification. The shape of the road plane reflects the natural road surface by implementing hybrid road plane that consists of rough road plane and dedicated road plane. The real-time implementation is possible due to the time efficient grouping and labelling. Even obstacle detection is achieved on the depth image. The error of obstacle detection due to the change in height and position of the road plane by pitch angle variation is corrected by the proposed hybrid road plane.