Layered Deformotion with Radiance: A Model for Appearance, Segmentation, Registration, and Tracking

Appearance models;Tracking;Registration;Segmentation;Image processing;Computer vision

Jackson, Jeremy D. ; Electrical and Computer Engineering

University:Georgia Institute of Technology

Department:Electrical and Computer Engineering

关键词: Appearance models;    Tracking;    Registration;    Segmentation;    Image processing;    Computer vision;   

Others  :  https://smartech.gatech.edu/bitstream/1853/16305/1/jackson_jeremy_d_200708_phd.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

This dissertation gives a general model for the estimation ofshape (image segmentation), appearance, pose (image registration), andmovement (tracking).The model can infer parameters formultiple objects in a dynamically changing scene. There are a number of real-world applications.In particular, in visual tracking, moving the camera to keep objects of interest in the field of view maycause the background to move.The objects canmove and deform in three dimensions, but they must be captured intwo-dimensional images.Each component of the image is represented bya separate layer: one for the background and a layer foreach foreground object.Each layer has three components: a contour that boundsthe region of the layer, a smooth function that represents the object'sappearance, and a transformation that maps that layer into an image.The segmentation for each layer is a contour (embedded as the zero level set of a distance function) that is the average shape of the object computed from multiple images.Thesmooth function associated with a layer approximates the image data inside thecontour, after the contour has been mapped into the image by asimilarity transformation (rigid component) plus a vector field (non-rigidcomponent).A practical application of having this model is thatone can fix the size of a layer and then construct priors on both shape and appearance for that layer.These priors areconstructed using principal components analysis (PCA), which reduces the dimensionality of theimage-approximating smooth function and the vector field (non-rigidregistration) and allows for more accurate modeling of an objectfor that layer.