|
Several image processing and computer vision tasks can be
attributed to quadrilateral detection. Examples include automatic
cropping of scanned photos, perspective correction of documents
acquired with digital cameras.
The major challenge is to achieve a robust performance even in
situations of low signal/noise ratio and correlated line-and-alike
noise.
Core Technology
We have developed a quadrilateral detection algorithm based on a
novel color edge detector:
-
a novel color edge detector that
boosts signal/noise ratio by taking local gradient orientation
into account
-
a line detector fully utilizes both
gradient strength and orientation
-
a detection scheme based on a metric maximization on sets of
mutually non-overlapping quadrilaterals
Applications I: Automatic perspective correction and
cropping of digital camera acquired document images
Assuming a complete document page is included in the image and a
reliable detection of the quadrilateral boundary as well as a known
aspect ratio, perspective correction and cropping can be done
automatically. No assumption about the document content is
necessary.
Here is an example (all images displayed as JPEG compressed PDF):
A magazine page acquired with a HP935 DC
Auto perspective corrected and
cropped
Adaptively enhanced
Application II: Automatic cropping of scanned photos
The constraints of corner angles closed to 90 degree, as well as
possibly common photo sizes, could be imposed to increase
reliability. No assumption about the number of photos as well
as their relative positions is necessary.
Here is an example:
Photos scanned with a HP see-thru
scanner
Photos detected
Contact
For more information about the technology, please contact Jian
Fan (jian.fan@hp.com) at
Imaging Technology Dept., HP Labs.
|
Printable version
