Files in this item
|(no description provided)|
|Title:||A unified scheme for image segmentation and object recognition|
|Author(s):||Qian, Richard Junqiang|
|Doctoral Committee Chair(s):||Huang, Thomas S.|
|Department / Program:||Electrical and Computer Engineering|
|Discipline:||Electrical and Computer Engineering|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
|Subject(s):||Engineering, Electronics and Electrical
|Abstract:||In this thesis, we present a unified scheme for image segmentation and object recognition. The scheme unifies image segmentation and object recognition via three serial stages: (1) optimal edge detection and region detection; (2) scale, position and orientation invariant object detection; and (3) high-level knowledge-based image segmentation.
In the first stage, edges are detected using a new edge detection algorithm and regions are extracted using a morphological-watershed-like segmentation algorithm. The new edge detection algorithm detects edges in an image with a curve-segment-based edge detection functional, which uses the Laplacian of Gaussian (LOG) zero-crossing contours as initial conditions to approach the true edge locations. The edge detection functional is shown to be optimal in terms of signal-to-noise ratio and edge localization accuracy for detecting general 2D edges. In addition, the resulting edge candidates preserve the nice scaling behavior of the LOG zero-crossing contours in scale space.
Based on the edge and region detection results from the first stage, the second stage detects objects of interest, e.g., human faces, using a new object detection algorithm. The new algorithm combines template matching methods with feature-based methods via hierarchical Markov random fields (MRFs) and maximum a posteriori probability (MAP) estimation. Hierarchical MRFs and MAP estimation provide a flexible framework to incorporate various visual clues. The combination of template matching and feature detection is shown to provide robustness for object detection against pose changes, complex background, and partial occlusions.
Finally based on the object detection results from the second stage, the third stage extracts the features of the detected targets, e.g., the eyes and the mouths of the detected faces, using a new deformable template matching algorithm. To reduce the probability of getting stuck in local minima, the new algorithm employs a coarse-to-fine scale space technique and uses the normalized cross-correlation to provide initial conditions for its deformation process. It also uses the optimal edge detection functional developed in this thesis to achieve the best accuracy for localizing the feature boundaries. Experimental results on real images from all the three stages are given in the thesis.
|Rights Information:||Copyright 1996 Qian, Richard Junqiang|
|Date Available in IDEALS:||2011-05-07|
|Identifier in Online Catalog:||AAI9712409|
This item appears in the following Collection(s)
Graduate Dissertations and Theses at Illinois
Graduate Theses and Dissertations at Illinois
Dissertations and Theses - Electrical and Computer Engineering
Dissertations and Theses in Electrical and Computer Engineering