Catálogo de publicaciones - libros

AdaBoost is a practical method of real-time face detection, but abides by a crucial problem of overfitting for the big number of features used in a trained classifier due to the weak discriminative abilities of these features. This paper proposes a theoretical approach to construct highly discriminative features, which is named composed features, from Haar-like features. Both of the composed and Haar-like features are employed to train a multi-view face detector. The primary experiments show promising results in reducing the number of features used in a classifier, which leads to the increase of the generalization ability of the classifier.

We propose a robust method to extract silhouettes of foreground objects from color video sequences. To cope with various changes in the background, the background is modeled as generalized Gaussian Family of distributions and updated by the selective running average and static pixel observation. All pixels in the input video image are classified into four initial regions using background subtraction with multiple thresholds, after which shadow regions are eliminated using color components. The final foreground silhouette is extracted by refining the initial region using morphological processes. We have verified that the proposed algorithm works very well in various background and foreground situations through experiments.

In dynamic scenes with occluding objects many features need to be tracked for a robust real-time camera pose estimation. An open problem is that tracking too many features has a negative effect on the real-time capability of a tracking approach. This paper proposes a method for the feature management which performs a statistical analysis of the ability to track a feature and then uses only those features which are very likely to be tracked from a current camera position. Thereby a large set of features in different scales is created, where every feature holds a probability distribution of camera positions from which the feature can be tracked successfully. As only the feature points with the highest probability are used in the tracking step, the method can handle a large amount of features in different scale without losing the ability of real time performance. Both the statistical analysis and the reconstruction of the features’ 3D coordinates are performed online during the tracking and no preprocessing step is needed.

Ukiyo-e is one famous traditional woodblock type Japanese drawing. Some pattern printed by special print techniques can only be seen from some special direction. This phenomenon relate to the reflection properties on the surface of Ukiyo-e. In this paper, we propose a method to measure these reflection properties of Ukiyo-e. Fitstly, the normal on the surface and the direction of the fiber in Japanese paper are computed from photos which are taken by a measuring machine named OGM. Then, fit the reflection model to the measured data and the reflection properties of Ukiyo-e can be obtained. Based on these parameters, the the appearance of Ukiyo-e can be rendered on real-time.

This paper proposes a novel and efficient feature-point matching algorithm for finding point correspondences between two uncalibrated images. The striking feature of the proposed algorithm is that the algorithm is based on the motion coherence/smoothness constraint only, which states that neighboring features in an image tend to move coherently. In the algorithm, the correspondences of feature points in a neighborhood are collectively determined in a way such that the smoothness of the local motion field is maximized. The smoothness constraint does not rely on any image feature, and is self-contained in the motion field. It is robust to the camera motion, scene structure, illumination, etc. This makes the proposed algorithm texture-independent and robust. Experimental results show that the proposed method outperforms existing methods for feature-point tracking in image sequences.

This paper proposes a new retrieval problem and conducts the initial study. This problem aims at finding the location of an item in a supermarket by means of visual retrieval. It is modelled as object-based retrieval and approached using the local invariant features. Two existing retrieval methods are investigated and their similarity measures are modified to better fit this new problem. More importantly, through the study this new retrieval problem proves itself to be a challenging task. An instant application of it is to help the customer find what they want without physically wandering around the shelves but a wide range of potential applications could be expected.

Iterative smoothing algorithms are frequently applied in image restoration tasks. The result depends crucially on the optimal stopping (scale selection) criteria. An attempt is made towards the unification of the two frequently applied model selection ideas: (i) the earliest time when the ‘entropy of the signal’ reaches its steady state, suggested by J. Sporring and J. Weickert (1999), and (ii) the time of the minimal ‘correlation’ between the diffusion outcome and the noise estimate, investigated by P. Mrázek and M. Navara (2003). It is shown that both ideas are particular cases of the marginal likelihood inference. Better entropy measures are discovered and their connection to the generalized signal-to-noise ratio is emphasized.

This paper proposes a general Kernel-Bayesian framework for object tracking. In this framework, the kernel based method—mean shift algorithm is embedded into the Bayesian framework seamlessly to provide a heuristic prior information to the state transition model, aiming at effectively alleviating the heavy computational load and avoiding sample degeneracy suffered by the conventional Bayesian trackers. Moreover, the tracked object is characterized by a spatial-constraint MOG (Mixture of Gaussians) based appearance model, which is shown more discriminative than the traditional MOG based appearance model. Meantime, a novel selective updating technique for the appearance model is developed to accommodate the changes in both appearance and illumination. Experimental results demonstrate that, compared with Bayesian and kernel based tracking frameworks, the proposed algorithm is more efficient and effective.

Object tracking using active contours has attracted increasing interest in recent years due to acquisition of effective shape descriptions. In this paper, an object tracking method based on level sets using moving cameras is proposed. We develop an automatic contour initialization method based on optical flow detection. A Markov Random Field (MRF)-like model measuring the correlations between neighboring pixels is added to improve the general region-based level sets speed model. The experimental results on several real video sequences show that our method successfully tracks objects despite object scale changes, motion blur, background disturbance, and gets smoother and more accurate results than the current region-based method.

In this paper, we present a new solution to the problem of multi-camera tracking with non-overlapping fields of view. The identities of moving objects are maintained when they are traveling from one camera to another. Appearance information and spatio-temporal information are explored and combined in a maximum a posteriori (MAP) framework. In computing appearance probability, a two-layered histogram representation is proposed to incorporate spatial information of objects. Diffusion distance is employed to histogram matching to compensate for illumination changes and camera distortions. In deriving spatio-temporal probability, transition time distribution between each pair of entry zone and exit zone is modeled as a mixture of Gaussian distributions. Experimental results demonstrate the effectiveness of the proposed method.