Kinh Tieu

Boosting Image Retrieval

We present an approach for image retrieval using a very large number of highly selective features and efficient learning of queries. Our approach is predicated on the assumption that each image is generated by a sparse set of visual “causes” and that images which are visually similar share causes. We propose a mechanism for computing a very large number of highly selective features which capture some aspects of this causal structure (in our implementation there are over 46,000 highly selective features). At query time a user selects a few example images, and the AdaBoost algorithm is used to learn a classification function which depends on a small number of the most appropriate features. This yields a highly efficient classification function. In addition we show that the AdaBoost framework provides a natural mechanism for the incorporation of relevance feedback. Finally we show results on a wide variety of image queries.

Fully automatic pose-invariant face recognition via 3D pose normalization

An ideal approach to the problem of pose-invariant face recognition would handle continuous pose variations, would not be database specific, and would achieve high accuracy without any manual intervention. Most of the existing approaches fail to match one or more of these goals. In this paper, we present a fully automatic system for pose-invariant face recognition that not only meets these requirements but also outperforms other comparable methods. We propose a 3D pose normalization method that is completely automatic and leverages the accurate 2D facial feature points found by the system. The current system can handle 3D pose variation up to ±45° in yaw and ±30° in pitch angles. Recognition experiments were conducted on the USF 3D, Multi-PIE, CMU-PIE, FERET, and FacePix databases. Our system not only shows excellent generalization by achieving high accuracy on all 5 databases but also outperforms other methods convincingly.

Inference of non-overlapping camera network topology by measuring statistical dependence

We present an approach for inferring the topology of a camera network by measuring statistical dependence between observations in different cameras. Two cameras are considered connected if objects seen departing in one camera is seen arriving in the other. This is captured by the degree of statistical dependence between the cameras. The nature of dependence is characterized by the distribution of observation transformations between cameras, such as departure to arrival transition times, and color appearance. We show how to measure statistical dependence when the correspondence between observations in different cameras is unknown. This is accomplished by non-parametric estimates of statistical dependence and Bayesian integration of the unknown correspondence. Our approach generalizes previous work which assumed restricted parametric transition distributions and only implicitly dealt with unknown correspondence. Results are shown on simulated and real data. We also describe a technique for learning the absolute locations of the cameras with Global Positioning System (GPS) side information

Automated multi-camera planar tracking correspondence modeling

This paper introduces a method for robustly estimating a planar tracking correspondence model (TCM) for a large camera network directly from tracking data and for employing said model to reliably track objects through multiple cameras. By exploiting the unique characteristics of tracking data, our method can reliably estimate a planar TCM in large environments covered by many cameras. It is robust to scenes with multiple simultaneously moving objects and limited visual overlap between the cameras. Our method introduces the capability of automatic calibration of large camera networks in which the topology of camera overlap is unknown and in which all cameras do not necessarily overlap. Quantitative results are shown for a five camera network in which the topology is not specified.

Learning semantic scene models by trajectory analysis

In this paper, we describe an unsupervised learning framework to segment a scene into semantic regions and to build semantic scene models from long-term observations of moving objects in the scene. First, we introduce two novel similarity measures for comparing trajectories in far-field visual surveillance. The measures simultaneously compare the spatial distribution of trajectories and other attributes, such as velocity and object size, along the trajectories. They also provide a comparison confidence measure which indicates how well the measured image-based similarity approximates true physical similarity. We also introduce novel clustering algorithms which use both similarity and comparison confidence. Based on the proposed similarity measures and clustering methods, a framework to learn semantic scene models by trajectory analysis is developed. Trajectories are first clustered into vehicles and pedestrians, and then further grouped based on spatial and velocity distributions. Different trajectory clusters represent different activities. The geometric and statistical models of structures in the scene, such as roads, walk paths, sources and sinks, are automatically learned from the trajectory clusters. Abnormal activities are detected using the semantic scene models. The system is robust to low-level tracking errors.

Correspondence-free multi-camera activity analysis and scene modeling

We propose a novel approach for activity analysis in multiple synchronized but uncalibrated static camera views. We assume that the topology of camera views is unknown and quite arbitrary, the fields of views covered by these cameras may have no overlap or any amount of overlap, and objects may move on different ground planes. Using low-level cues, objects are tracked in each of the camera views independently, and the positions and velocities of objects along trajectories are computed as features. Under a generative model, our approach jointly learns the distribution of an activity in the feature spaces of different camera views. It accomplishes two tasks: (1) grouping trajectories in different camera views belonging to the same activity into one cluster; (2) modeling paths commonly taken by objects across camera views. To our knowledge, no prior result of co-clustering trajectories in multiple camera views has been published. Advantages of this approach are that it does not require first solving the challenging correspondence problem, and the learning is unsupervised. Our approach is evaluated on two very large data sets with 22, 951 and 14, 985 trajectories.

Estimation of Signal Information Content for Classification

Information measures have long been studied in the context of hypothesis testing leading to variety of bounds on performance based on the information content of a signal or the divergence between distributions. Here we consider the problem of estimation of information content for high-dimensional signals for purposes of classification. Direct estimation of information for high-dimensional signals is generally not tractable therefore we consider an extension to a method first suggested in [1] in which high dimensional signals are mapped to lower dimensional feature spaces yielding lower bounds on information content. We develop an affine-invariant gradient method and examine the utility of the resulting estimates for predicting classification performance empirically.

Estimating dependency and significance for high-dimensional data

Understanding the dependency structure of a set of variables is a key component in various signal processing applications which involve data association. The simple task of detecting whether any dependency exists is particularly difficult when models of the data are unknown or difficult to characterize because of high-dimensional measurements. We review the use of nonparametric tests for characterizing dependency and how to carry out these tests with high-dimensional observations. In addition we present a method to assess the significance of the tests.

Exploiting user labels with generalized distance transforms random field level sets

We present an approach for exploiting user labels with random field level sets in image segmentation. A sparse set of user labels is propagated to the rest of the image by computing a generalized distance transform which takes into account image intensity information. The region-based level set formulation is modified to use random field level sets whose range is restricted to the probability values. These two ideas are combined in a single level set functional. Improved results are shown on a liver segmentation task.