Next: The immersive environment Up:
Title Page Previous: Full
text index Index: Full text index
Contents: Conference Page
Introduction
Segmentation can only occur where a precise description
of structure exists. The language for describing image structure includes:
thresholds on a scalar data value range (e.g. x-ray opacity), regions within
a vector space of data values (e.g. RGB space), regions within a texture
statistics space, gradient edges, transforms for isolating lines and parametric
curves, and geometric arrangements of primitive shapes. There is a wide
gap between this language and descriptions of anatomically or medically
significant features. Humans bring a great deal of expert and contextual
knowledge to bear on image recognition problems. When presented with such
recognition problems they mentally engage in an iterative visual refinement
as they focus on the important features of the image data. This refinement
involves a sequence of choices which seek to distinguish significant visual
details from irrelevant information. These choices are made in parallel
with a visual exploration of the image. This parallel activity is important
since the appropriate choices for segmentation are frequently not known
until the data has been partially visualized.
This paper describes the first step of a project to exploit the expert
interpretive knowledge of users by providing an interactive, immersive
environment in which users directly control the segmentation and visualization
process. The Visible Human cryosection data set offers a unique test set
in which to examine these ideas.
Next: The immersive environment Up:
Title Page Previous: Full
text index Index: Full text index
Contents: Conference Page
