by M Wimmer and B Radig
Abstract:
Model-based image interpretation has proven to appropriately extract high-level information from images. A priori knowledge about the object of interest represents the basis of this task. Model fitting determines the model that best matches a given image by searching for the global optimum of an objective function. Unfortunately, the objective function is usually designed manually, based on implicit and domain-dependent knowledge. In contrast, this paper describes how to obtain highly accurate objective functions by learning them from annotated training images. It automates many critical decisions and the remaining manual steps hardly require domain-dependent knowledge at all. This approach yields highly accurate objective functions. Our evaluation fits a face model to a publicly available image database and compares the obtained results to a recent state-of-the-art approach.
Reference:
Automatically Learning the Objective Function for Model Fitting (M Wimmer and B Radig), In Proceedings of the Meeting in Image Recognition and Understanding (MIRU), 2007.
Bibtex Entry:
@inproceedings{wimmer_automatically_2007,
author = {M Wimmer and B Radig},
title = {Automatically Learning the Objective Function for Model Fitting},
booktitle = {Proceedings of the Meeting in Image Recognition and Understanding
({MIRU)}},
year = {2007},
address = {Hiroshima, Japan},
month = {jul},
abstract = {Model-based image interpretation has proven to appropriately extract
high-level information from images. A priori knowledge about the
object of interest represents the basis of this task. Model fitting
determines the model that best matches a given image by searching
for the global optimum of an objective function. Unfortunately, the
objective function is usually designed manually, based on implicit
and domain-dependent knowledge. In contrast, this paper describes
how to obtain highly accurate objective functions by learning them
from annotated training images. It automates many critical decisions
and the remaining manual steps hardly require domain-dependent knowledge
at all. This approach yields highly accurate objective functions.
Our evaluation fits a face model to a publicly available image database
and compares the obtained results to a recent state-of-the-art approach.},
}