image restoration for halftone pattern printed pictures in...
TRANSCRIPT
Image Restoration for
Halftone Pattern Printed Pictures
in Old Books
Adrian Ciobanu, Tudor Barbu, Mihaela Luca
Institute of Computer Science, Romanian Academy – Iasi Branch
Conventional-dot halftoningused in Romanian old books dated arround 1935
for printing images
Digital halftoningintroduced in 1970, based on
printing technology advances
Scanned pictures and detailsat 600 dots per inch
Pictures and detailsin a PDF available version
Proposed restoring procedure
• Scan a picture from an original book, preferably with the highest resolution
possible and in grayscale.
• Find a global optimum threshold and transform the scanned image in black and
white. The result is a collection of black spots on a white background or white
spots on a black background. These spots are centered on a grid specific to each
halftone pattern method.
• Preprocess the thresholded image by detecting and deleting isolated black pixels
on white background.
• Detect all the black spots on white background with less than 20 pixels in
composition and transform them in black spots with regular shape (close to a
disk) depending on their size. In this way the analogue halftoning is converted in a
digital halftoning. Black spots with more than 19 pixels will be addressed at a later
step.
• Detect all the white spots on black background with less than 20 pixels in
composition and transform them in white spots with regular shape depending on
their size. White spots with more than 19 pixels will be addressed at a later step.
Proposed restoring procedure
• Based on the obtained regular black spots with less than 20 pixels in composition,
detect the grid points of the original halftone pattern as their center of mass or
centroid. Apply a recurrent procedure to fill as much as possible the missing grid
points if three already known neighbor grid points are available.
• For black spots with more than 19 pixels apply cleaning techniques to divide them
in several black spots with less than 20 pixels in composition, using as context
information the already detected grid points. Then apply a regularization of shape
also for them.
• A similar technique should be applied for white spots with more than 19 pixels
(rarer and not so important for the quality of the restored image).
• The final restored halftone pattern image will contain only regular black and white
digital spots, with leaks between them almost completely reduced, so the restored
image will look very clean.
• The last step will consist of an inverse halftoning procedure that will produce a
gray level image with superior quality, ready to be used in common everyday
tasks.
Implementationgraylevel image → black&white image
optimal Otsu threshold
Implementationblack&white image → mostly black & mostly white regions
Implementationmostly white regions mostly black regions
ImplementationPreprocessing of mostly white regions
initial black&white image → eliminating pixels on edges
→ eliminating several types of connecting pixels
ImplementationReplacing analogue halftoning with digital halftoning
Black spots with 3 to 6 pixels
ImplementationReplacing analogue halftoning with digital halftoning
Black spots with 7 to 10 pixels
ImplementationReplacing analogue halftoning with digital halftoning
Black spots with 11 to 14 pixels
ImplementationReplacing analogue halftoning with digital halftoning
Black spots with 15 to 19 pixels
ImplementationReplacing analogue halftoning with digital halftoning
Final cleaning of isolated 1 to 3 pixels spots
ImplementationPreprocessing of mostly black regions
black&white image after processing mostly white regions
→ eliminating isolated pixels
ImplementationReplacing analogue halftoning with digital halftoning
White spots with 3 to 6 pixels – formation I
ImplementationReplacing analogue halftoning with digital halftoning
White spots with 3 to 6 pixels – formation II
ImplementationReplacing analogue halftoning with digital halftoning
White spots with 7 to 10 pixels
ImplementationReplacing analogue halftoning with digital halftoning
White spots with 11 to 14 pixels
ImplementationReplacing analogue halftoning with digital halftoning
White spots with 15 to 19 pixels
ImplementationReplacing analogue halftoning with digital halftoning
Final cleaning of isolated 1 to 3 pixels spots
Implementationgraylevel image → 50,35% processed black&white image
Other resultsgraylevel image → 46,14% processed black&white image
Other resultsgraylevel image → 37,25% processed black&white image
Other resultsgraylevel image → 42,43% processed black&white image
Other resultsgraylevel image → 36,22% processed black&white image
Other resultsgraylevel image → 44,97% processed black&white image
Other resultsgraylevel image → 38,16% processed black&white image
Other resultsgraylevel image → 45,62% processed black&white image
Other resultsgraylevel image → 33,18% processed black&white image
Other resultsgraylevel image → 40,59% processed black&white image
Other resultsgraylevel image → 39,13% processed black&white image
Conclusions and further work
• A novel halftoning image restoration algorithm for pictures printed in old books
has been proposed.
• Its implementation is now in the stage of completely and successfully processing
black and white spots with less than 20 pixels in composition, replacing them with
digital regular shapes, which makes the image clearer and cleaner.
• Further work will imply implementing the cleaning of black and white spots with
more than 20 pixels in composition, based on information extracted about the grid
of halftoning points.
• This will completely clean the image and restore it almost perfectly to the halftone
image that the publishing house wanted to print in the book.
• The last stage should implement an inverse halftoning technique meant to give
the image the feeling of nowadays graylevel images and make it ready to be used
in any computer environment.
Thank you very much!