[Oradlist] Image processing with digital imaging
rmolteni at afpimaging.com
Sat Jul 5 01:19:59 PDT 2008
Verona, 2008 July 5th
I concur 110% with what Alan Farman and Douglas Benn have already responded on this subject (and about Analoui's articles being an excellent reference).
There isn’t – or it is extremely hard to define - such a thing as a “raw image” (or "raw data" when we speak of 3D volumetric imaging as in CBCT); and do not hope that manufactures will be much eager to reveal the proprietary know-how (a combination hardware, firmware, and software) with which they get to present the “best possible” image to the operator – in part, also because they may not fully know it themselves, as I will further explain below.
However, there are certain general comments that I can offer, which are applicable to all digital radiography and to Cone Beam Volumetric Imaging in particular, irrespective of the make.
Some pre-processing is performed directly at level of the software driver for the digital image detector itself, and are involve and intimate commixture of HW and SW functions. That often include correction for defective pixels or blemished in the detector/camera, whose signal contribution is eliminated and replaced by that of an adjacent pixel (in the simplest technique) or by the average of the surrounding pixels (in more advanced techniques); and correction for the individual gain and offset variations in the detector array. This is done by means of a “masking file” which associated to each specific detector and that can be saved either in some internal memory of the detector, or in the memory of the computer driving that system. In the latter case, if such masking file is lost or corrupted (or if computers from different systems are swapped) the images may get flawed early in the pre-processing chain.
This kind of correction may even happen twice in the overall preprocessing chain, i.e. first strictly related to just the detector itself, and quite transparent to the user and sometimes even to the overall system designer, and subsequently to account also for various other factors in the imaging system (e.g. x-ray field non-uniformity) in which case the process is less transparent (blank field acquisition, etc.).
In case of volumetric digital radiography (e.g. CBCT), another kind of pre-processing is the application of geometrical correction factors during the primary reconstruction (e.g. with a Feldkamp or ART algorithm) from the projection images to the volumetric dataset (3D), that are specific for the individual system and account for minor (minuscule, sometimes) departures from its nominal geometrical construction. Such factors are (automatically) calculated by performing geometrical calibration protocols by means of proprietary-designed calibration tools.
Then we have transformations in the gray scale domain. The initial data are often captured at, say, 12 or 14 bits (and padded into a 16 bit file format) and must be fit into a displayable 8 bit format. Often this involves identifying the useful grayscale range (i.e. the part of the whole 16 bit range that actually contains meaningful information) and “linearly stretching” it into the displayed 8 bit (usually by cropping away the parts in the initial histogram below a certain percentile (say, 1%).
Or the stretching may not be linear, but via a “shaped” LUT (Look-Up Table); that is, the relationship between the initial data and the displayed one may not be a straight line but a curve that is optimized for the specific diagnostic task: different LUTs are better suitable for, say, a lateral projection of the human head for orthodontic cephalometrics, or the breast for 3D mammography, or the hoof of and horse! This kind of pre-processing may or may not be transparent to the operator.
Another class of pre-processing relates to spatial frequency filtering, whereas certain filters may be (automatically and transparently) applied to selectively remove noise (such as quantum noise) without too much deteriorating detail resolution – or, conversely, to enhance edge visibility without too much impairing signal-to-noise-ratio.
I apologize for belated response to this inquiry. I have been extremely busy at working at (transparent) pre-processing algorithms for our Cone Beam Volumetric Imaging machines and digital radiography systems.
Chief Technical Officer
AFP Imaging / Dent-X / NewTomDental / QR
From: oradlist-bounces at lists.ucla.edu [oradlist-bounces at lists.ucla.edu] On Behalf Of Sharon Brooks [slbrooks at umich.edu]
Sent: Thursday, June 26, 2008 3:48 PM
Cc: Jeremy Kott
Subject: [Oradlist] Image processing with digital imaging
A graduate student in endodontics is working on a thesis project related to
digital imaging. He posed some questions for me regarding pre-processing of
digital images before they appear on the screen. I'm pretty sure that
everything is processed somehow before you see it, but this is definitely
not my strong area. Therefore, I am asking for help for him.
I am posting his question to me. Any assistance would be appreciated.
********** Question from graduate student ***********
I was wondering if there is any literature/information out there on what
sort of processing the digital images go through before they show up on the
screen? In several of the software packages, you get the unaltered image,
but then there is an image enhancement button which can be pressed to
"enhance" the image. I was wondering what that actually does, is it some
sort of algorithm? And if it does that, are we really seeing the true image
on the screen. Also, with Optime PSP images, they don't seem to have that
function. You can manipulate the images with contrast, brightness and the
like, but is the image which comes on the screen the true image or does that
go through processing as well? I was hoping you could point me in the
direction of some literature which might explain this to me.
Sharon L. Brooks, DDS, MS
Diplomate, American Board of Oral and Maxillofacial Radiology
University of Michigan School of Dentistry
Department of Periodontics and Oral Medicine
Ann Arbor, MI 48109-1078 USA
Tel: +1 734-764-1595 Fax +1-734-764-2469
slbrooks at umich.edu
Oradlist mailing list
Oradlist at lists.ucla.edu
More information about the Oradlist