We present an unsupervised technique for visual learning which is based on density estimation in high-dimensional spaces using an eigenspace decomposition. Two types of density estimates are derived for modeling the training data: a multivariate Gaussian (for unimodal distributions) and a Mixture-of-Gaussians model (for multimodal distributions). These probability densities are then used to formulate a maximum-likelihood estimation framework for visual search and target detection for automatic object recognition and coding. Our learning technique is applied to the probabilistic visual modeling, detection, recognition, and coding of human faces and non-rigid objects such as hands.

Weinvestigate the use of linear and nonlinear principal manifolds for learning lowdimensional representations for visual recognition. Three techniques: Principal Component Analysis #PCA#, Independent Component Analysis #ICA# and Nonlinear PCA #NLPCA# are examined and tested in a visual recognition experiment using a large gallery of facial images from the #FERET" database. We compare the recognition performance of a nearest-neighbour matching rule with each principal manifold representation to that of a maximum aposteriori #MAP# matching rule using a Bayesian similarity measure derived from probabilistic subspaces and demonstrate the superiority of the latter.

An active object recognition system is built using three different uncertainty calculi: probability theory, possibility theory (or fuzzy logic) and Dempster-Shafer theory of evidence. The system is allowed to reposition the camera to capture additional views and, therefore, to resolve the classification result obtained from a single view. The approach uses an appearance based object representation, namely the parametric eigenspace, but the action planning steps are largely independent of any details of the specific object recognition environment. The active recognition problem can be tackled succesfully by all three approaches with sometimes only slight differences in performance. The results obtained in extensive test runs confirm that recognition rate can be improved considerably by performing active steps. Random selection of the next action is much less efficient than planning, both in recognition rate and in the average number of steps required for recognition. Remar...

We present an efficient method within an active vision framework for recognizing objects which are ambiguous from certain viewpoints. The system is allowed to reposition the camera to capture additional views and, therefore, to resolve the classification result obtained from a single view. The approach uses an appearance based object representation, namely the parametric eigenspace, and augments it by probability distributions. This captures possible variations in the input images due to errors in the pre-processing chain or the imaging system. Furthermore, the use of probability distributions gives us a gauge to view planning. View planning is shown to be of great use in reducing the number of images to be captured when compared to a random strategy. 1 Introduction Most computer vision systems found in the literature perform object recognition on the basis of the information gathered from a single image. Typically, a set of features is extracted and matched against object ...

We present an efficient method within an active vision framework for recognizing objects which are ambiguous from certain viewpoints. The system is allowed to reposition the camera to capture additional views and, therefore, to improve the classification result obtained from a single view. The approach uses an appearance-based object representation, namely the parametric eigenspace, and augments it by probability distributions. This enables us to cope with possible variations in the input images due to errors in the pre-processing chain or changing imaging conditions. Furthermore, the use of probability distributions gives us a gauge to perform view planning. Multiple observations lead to a significant increase in recognition rate. Action planning is shown to be of great use in reducing the number of images necessary to achieve a certain recognition performance when compared to a random strategy. # 2000 Elsevier Science B.V. All rights reserved. Keywords: Action planning

Colour is one of the most important features in content based image retrieval. However, colour is rarely used as a feature that codes local spatial information, except for colour texture. This paper presents an approach to represent spatial colour distributions using local principal component analysis (PCA). The representation is based on image windows which are selected by two complementary data driven attentive mechanisms: A symmetry based saliency map and an edge and corner detector. The eigenvectors obtained from local PCA of the selected windows form colour patterns that capture both low and high spatial frequencies, so they are well suited for shape as well as texture representation. Projections of the windows selected from the image database to the local PCs serve as a compact representation for the search database. Queries are formulated by specifying windows within query images. System feedback makes both the search process and the results comprehensible for the user.

The SLAM software package has been developed for appearance learning and matching problems in computational vision. Appearance learning involves use of principal component analysis for compression of a large input image set to a compact low-dimensional subspace, called the eigenspace, in which the images reside as parameterized manifolds. SLAM enables the user to obtain this parametric representation by providing modules for eigenspace computation, projection of images to eigenspace, and interpolation of multivariate manifolds through the projections. Appearance matching is done by searching for a projection in eigenspace closest to a novel input projection. Algorithms have been provided for performing this search in real-time, even with huge datasets. Benchmarks demonstrate the suitability of SLAM for application to real-world problems. The functionality has been made available to the user through an X/Motif Graphical User Interface along with commandline programs and a C++ class library. Use of object oriented techniques provides an easy to use and extensible Application Programming Interface. 1

This is an overview of image understanding research at Columbia's Center for Research in Intelligent Systems since the 1993 IU workshop. It reviews our work on the following topics: 1. Physical Models and Sensors (a) Seeing Beyond Lambert's Law (b) Impossible Shape from Shading (c) Specular Reflection and Stereo Vision (d) Color and Polarization (e) Active Shape from Focus 2. Visual Learning and Recognition (a) Learning Object Appearance Models (b) Real-Time Object Recognition System (c) Software Library for Appearance Matching 3. Sensor and Illumination Planning (a) Illumination Planning for Recognition (b) Model Based Sensor Planning 4. Shape Modeling (a) Extruded Generalized Cylinders (b) Periodic Generalized Cylinders (c) Deformable Object Models 5. Vision for Robotics (a) Positioning and Tracking Appearance (b) Visual Control for Grasping (c) Automatic CAD Model Acquisition (d) Uncertainity Models for Robotic Tasks 6. Qualitative Vision (a) Uncalibrated Vis...

The SLAM software package has been developed for appearance learning and matching problems in computational vision. Appearance learning involves use of principal component analysis for compression of a large input image set to a compact low-dimensional subspace, called the eigenspace, in which the images reside as parameterized manifolds. SLAM enables the user to obtain this parametric representation by providing modules for eigenspace computation, projection of images to eigenspace, and interpolation of multivariate manifolds through the projections. Appearance matching is done by searching for a projection in eigenspace closest to a novel input projection. Algorithms have been provided for performing this search in real-time, even with huge datasets. Benchmarks demonstrate the suitability of SLAM for application to real-world problems. The functionality has been made available to the user through an X/Motif Graphical User Interface along with commandline programs an...

We present an efficient method within an active vision framework for recognizing objects which are ambiguous from certain viewpoints. The system is allowed to reposition the camera to capture additional views and, therefore, to improve the classification result obtained from a single view. The approach uses an appearance based object representation, namely the parametric eigenspace, and augments it by probability distributions. This enables us to cope with possible variations in the input images due to errors in the pre-processing chain or changing imaging conditions. Furthermore, the use of probability distributions gives us a gauge to perform view planning. Multiple observations lead to a significant increase in recognition rate. Action planning is shown to be of great use in reducing the number of images necessary to achieve a certain recognition performance when compared to a random strategy. 1 Introduction Most computer vision systems found in the literature perform object recog...