26-07-2012, 10:18 AM
User-Friendly Interactive Image Segmentation Through Unified Combinatorial
User-Friendly Interactive Image Segmentation.pdf (Size: 874.8 KB / Downloads: 27)
Abstract
One weakness in the existing interactive image segmentation
algorithms is the lack of more intelligent ways to understand
the intention of user inputs. In this paper,we advocate the use
of multiple intuitive user inputs to better reflect a user’s intention.
In particular, we propose a constrained random walks algorithm
that facilitates the use of three types of user inputs: 1) foreground
and background seed input, 2) soft constraint input, and 3) hard
constraint input, as well as their combinations. The foreground
and background seed input allows a user to draw strokes to specify
foreground and background seeds. The soft constraint input allows
a user to draw strokes to indicate the region that the boundary
should pass through. The hard constraint input allows a user to
specify the pixels that the boundary must align with. Our proposed
method supports all three types of user inputs in one coherent computational
framework consisting of a constrained random walks
and a local editing algorithm, which allows more precise contour
refinement.
INTRODUCTION
I NTERACTIVE image segmentation involves minimal user
interaction to incorporate user intention into the segmentation
process and is an active research area in recent years because
it can achieve satisfactory segmentation results that are
unattainable by the state-of-the-art automatic image segmentation
algorithms. This paper considers the same problem of how
to interactively segment a foreground object out from its surrounding
background. Our goal is to develop intuitive and intelligent
image segmentation algorithms and tools that allow users
to interactively guide the segmentation algorithm via a small
amount of intuitive interactions until a satisfactory segmentation
result that reflects both user intentions and photometric features
is achieved.
Related Work
In general, interactive image segmentation can be classified
into two categories: hard segmentation and soft segmentation.
Hard segmentation algorithms such as [1], [2] produce a binary
map, i.e., a pixel belongs to either foreground or background,
while soft segmentation algorithms such as [3], [4] extract a
fractional (fuzzy) matte for an image. In this research, we only
consider the hard segmentation problem. In the following, we
give a brief review on the related interactive image segmentation
algorithms and tools.
Early interactive image segmentation algorithms utilize either
regional properties such as Adobe’s magic wand or boundary
properties such as active contour [5] and intelligent scissors [6],
[7]. The magic wand tool starts with a small user-specified region.
The region grows through connecting neighboring pixels
that fall within some adjustable tolerance range of the color statistics
of the specified region. With the active contour method,
the user is typically asked to place a contour near the desired
boundary and the algorithm evolves the boundary to snap to
the object contour. The main problem with the active contour
method is that the contour is likely to be trapped in a local minimum.
The intelligent scissors algorithm requires the user to
place points along the desired contour of the foreground object.
Dijkstra’s shortest path algorithm is used to compute the path
between neighboring points.
CONCLUSION
In this paper, we have proposed an interactive image segmentation
framework that consists of two components: constrained
random walks and local contour deformation. The proposed
framework supports multiple intuitive types of user inputs
and therefore combines the advantages of different user
interactions. The foreground and background brushes are the
most commonly used interaction tools as they are easy to use
and instructive to the algorithms. The soft boundary brush and
the hard boundary pixel selector are extremely useful to handle
weak boundaries, where adding more foreground or background
strokes may cause unexpected fluctuation in the segmentation
results. These tools enable the proposed framework to work fast
and accurately with ease. The superior performance of the algorithm
has been demonstrated by a number of experiments on
the benchmark data sets.