|
Free downloading
User's manual (in PDF)
|
Image masking
In video composition, a skilled user is often required to manually extract
objects from an arbitrary background of a video sequence and include them
in another sequence. This is an inaccurate and imprecise task that usually
takes half of the total time for video production. This process is commonly
known as image masking (or image segmentation).
ProntoMask
ProntoMask is a solution that considerably reduces the time spent by the
user in the image masking process. ProntoMask is part of a larger project
to study and develop image segmentation techniques for automatic and interactive
object delineation, extraction and insertion in static and dynamic images
(video). ProntoMask is a general tool, independent of application, that
can also be used as an image segmentation tool in medical imaging.
ProntoMask consists of two different masking tools: ProntoContour
and ProntoRegion. Currently, these tools work for still images.
ProntoContour
In ProntoContour, the user first selects an initial point on the boundary
of the object. As the user moves the mouse, a dynamic line connecting the
initial point and the cursor is displayed in realtime (Fig 1a). If the
cursor is near to the boundary, this line snaps onto it (Fig 1b). If the
line describes the object boundary appropriately, the user deposit the
cursor which now becomes the new starting point. The process continues
in this fashion (Fig 1c) until the user decides to close the contour (Fig
1d). Once the contour is closed, the user can save the mask associated
with the object.
|
|
|
(a)
|
(b)
|
|
|
|
(c)
|
(d)
|
Fig. 1 - ProntoContour
ProntoRegion
In ProntoRegion the user initially draws white lines inside the object
and black lines in the background (Fig 2a). ProntoRegion immediately responds
by displaying the masked object (Fig 2b). This result is computed by partitioning
the image into two regions. The object region (mask) with pixels most similar
to the pixels under the white lines and the background region with pixels
most similar to the pixels under the black lines. The user can add or remove
lines to improve this result. In the illustrative example below, by adding
a white line in the red shirt (Fig 2c), the mask grows to include the shirt
and trousers (Fig 2d). The process continues in this fashion until the
entire object is detected (Fig 2e-f).
|
|
|
(a)
|
(b) |
|
|
|
(c)
|
(d)
|
|
|
| (e) |
(f) |
Fig 2 - ProntoRegion
|