- Larry S. Davis - Publications

High Performance Computing

High Performance Computing for Land Cover Dynamics. This is an overview paper presented at the International Joint Conference on Pattern Recognition, Jerusalem, in October 1994.

High Performance Computing for Image Processing. This paper contains descriptions of two new parallel algorithms for image classification - one based on hierarchical Markov Random Fields (due to Rama Chellappa) and the second based on utilizing total probability models for classifying regions in a hierarchical segmentation of a remotely sensed image. The paper was presented at the first International Workshop on Parallel Processing in Bangalore, December 1994.

A High Performance Image Database System for Remotely Sensed Imagery. This paper contains the design of a high performance image database system for global image data sets obtained through remote sensing. Specific issues that arise in the organization of large amounts of AVHRR images, and the integration of maps and spatial queries, are discussed.

Parallel Algorithms for Image Enhancement and Segmentation by Region Growing with an Experimental Study. A description of a parallel and portable implementation of image segmentation using a hierarchical connected components algorithm.

Visual Navigation

Detection of Independently Moving Objects in Passive Video. Two different approaches for the identification of independently moving objects are presented and applied to video sequences taken from an autonomous vehicle.

An Improved Radial Basis Function Network for Autonomous Road-Following. A Radial Basis Function Architecture for road following is presented and compared, on a driving simulator, to a backpropagation road follower modeled after ALVINN. Several sets of experiments on the CMU NAVLAB are also presented. This paper will appear in Neural Networks

RSTA on the Move - Progress Report. This report, to appear in the 1996 Image Understanding Workshop, summarizes research done on our RSTA on the Move project during the previous year.

Appearance-based Vision

An Appearance Based Approach to Object Recognition in Aerial Images. This paper describes a system for recognition of objects in aerial images based on a heuristically controlled combinatorial search applied to a hierarchical segmentation of the image.

Heads, Faces and Facial Expressions

Recognizing Faces Showing Expressions. This paper contains a comparison of graph matching and eigneface approaches to face recognition, applied to a large database of faces showing a variety of facial expressions. The paper was presented at the International Workshop on Face and Gesture Recognition in Zurich, held in June 1995.

Human Emotion Recognition from Motion Using a Radial Basis Function Network Architecture. A radial basis function neural network architecture is developed which can learn the facial deformation patterns that characterize different facial expressions. This paper will appear in the IEEE Transactions on Neural Networks .

Computing Spatio-Temporal Representations of Human Faces. A version of the longer technical report referenced below that was presented at CVPR '94. This paper presents the basic system architecture for the system that recognizes facial expressions, and contains the results of some preliminary experiments.

Recognizing Facial Expressions by Spatio-Temporal Analysis. This short paper contains the results of an experimental study in which CAFE, our system for recognition of facial expressions, was applied to a database of 32 individual displaying six facial expressions.

Recognizing Facial Expressions. A Center for Automation Research technical report that contains a detailed description of the CAFE (Computer Analysis of Facial Expressions)system. This paper will appear in the IEEE Transactions on Pattern Analysis and Machine Intelligence.

Recognizing facial expressions in image sequences using local parameterized models of image motion. This paper uses parametric flow field models both to recover the rigid motion of the head and the nonrigid motion of face features. The parameters of the nonrigid motion are then used to recognize facial expressions. Examples are provided on both laboratory sequences of subjects and "live" video from TV and movies. [ This file is 13MB compressed postcript - beware! ]

Bodies - Surveillance of People in Action

3-D Model-based Recognition of Human Movement by Dynamic Time Warping. This paper addresses the problem of recognition of human movement patterns using a dynamic time warping approach. Moving light displays of humans performing simple actions (waving, etc.) are used to both acquire representations of these movements as sequences of instantaneous poses and as a basis for recognition.

Tracking of humans in action: a 3-D model-based approach. This paper describes an approach to recovering the time varying pose of a human body in action based on viewing the activity from several vantage points simultaneously. It appeared in the 1996 ARPA image Understanding Workshop.