| |
KF5JRV > TECH 29.06.16 12:23l 127 Lines 7131 Bytes #999 (0) @ WW
BID : 5215_KF5JRV
Read: GUEST
Subj: Earliest Image Processing
Path: IW8PGT<CX2SA<N0KFQ<KF5JRV
Sent: 160629/1116Z 5215@KF5JRV.#NWAR.AR.USA.NA BPQ1.4.65
Earliest Image Processing
By 1957 computers were in common use in many laboratories and
commercial establishments. Originally, they were devoted
exclusively to numerical, algebraic, and geometric computation.
Later, the symbol manipulation capability of computers became
recognized, leading to so-called business data processing in
which alphanumeric processing became routine. The alphanumeric
data presented an obvious problem of inputting the vast quantity
of data needed for business. This created activity in developing
character recognition machinery (9). It occurred to R.A.Kirsch
that a general purpose computer could be used to simulate the
many character recognition logics that were being proposed for
construction in hardware. This would require an input device
that could transform a picture (of a character) into a form
suitable for storage in the memory of a computer.
A further important advantage of building such a device was that
it would enable programs to be written to simulate the multifarious
ways in which humans view the visible world. A tradition had been
building in which simple models of human structure and function had
been studied, for example, in neuroanatomy and neurophysiology.
The emphasis on binary representations of neural functions led us
to believe that binary representations of images would be suitable
for computer input. This serious mistake, discussed below, was
implemented in the first picture scanner built. It was connected to
the SEAC in 1957 and it enabled Kirsch's group to experiment with
algorithms that launched the fields of image processing and image
pattern recognition.
The scanner used a rotating drum and a photomultiplier to sense
reflections from a small image mounted on the drum. A mask
interposed between the picture and the photomultiplier tessellated
the image into discrete pixels.
The group experimented with several classes of algorithms. The first
was homogeneous transformations. Once an image was acquired, the
great speed of SEAC was used to transform it with edge enhancement
filters. These have become important in recent years as highly
parallel methods of processing became common in neural network
simulations, for example. They also provided the basis for the
large class of image enhancement methods that developed. The group
also wrote algorithms to make measurements on objects in an image.
By showing that these objects could have multiple connectivity and
still be measured correctly, they encouraged the development of
specialized machines for image analysis.
A staticizer connected to the SEAC memory enabled a stored image
to be displayed on a cathode ray oscilloscope. This made it possible
for the researchers to see what the computer "saw". And when they
could see binary images, they realized the limitations of binary
representation. So they experimented with superimposing multiple
scans at different scanning thresholds and the use of time varying
thresholds for pulse density modulation to represent multiple gray
levels in an image.
While new applications of image processing were being developed at NBS,
new tools, both physical and conceptual, were being developed.
Reference stated that "the best way to store a photograph is in
its original form." This implied that for many photographs, a larger
scanner than the first SEAC scanner would be necessary if the
computer was to have access to such images. In 1964, R.T.Moore,
M.C.Stark, and L.Cahn at NBS built a precision scanner that could
accommodate a much larger image. This scanner was built around
a commercial lathe body offering dimensional precision of 0.127 mm.
It had sufficient accuracy that repeated scans could serve the
purpose of avoiding the prohibitively large memory requirements
that would be needed to store a scanned image as large as 250 mm
square.
In 1957, Kirsch et.al. thought that pattern recognition could proceed
monotonically forward from scanning to processing to recognition. It
was many years before they understood that models of the visual world
would have to exist in the computer before scanned images could be
effectively used. By 1964 Kirsch showed that one could summarize
the information about the visual world in the form of a picture grammar.
Such a grammar would precede the image scanning operation and aid in
pattern recognition. Much of this work was theoretical in nature and
resulted in a large literature on syntactic image processing. Only in
1978, did this understanding get reformulated elsewhere in the
context of architecture and applied to practical problems in
representation of architectural designs.
Among the tools developed were programming languages for processing
images. In Moore describes a subroutine library for the SEAC,
written by R. B. Thomas, that made it convenient to invoke processing
operations on metallurgical photograph images. Later, K. Kloss
adapted for the IBM 709 computer at NBS, a simulator written at the
University of Illinois. This simulator was for the Pattern
Articulation Unit of the Illiac 3 computer. It was a homogeneous
processor of the kind that became common with parallel processing.
Kloss embedded the program in the LISP language. NBS staff also
installed the system on the ANFSQ32 computer at System Development
Corp. in California and used it, again remotely on slow telephone
lines. D. J. Orser, I. Rhodes, and A. H. Meininger, all of NBS, also
developed a LISP based image processing system installed on the
DEC-10 computer at the National Institutes of Health that was used
remotely at NBS.
The most recent, powerful member of this class of LISP-based image
processing languages is the MacLispix language developed by David S.
Bright at NIST. MacLispix uses the Macintosh Common Lisp language
on the Macintosh computer. This is a LISP compiler and interpreter,
with the full symbol manipulation capability of LISP, in which Bright
has embedded efficiently coded routines for most common image
processing operations. It is available as a public domain language
with extensive documentation.
When NBS started the field of computer image processing, those principal
engineers and scientists involved in computer image processing could
not have anticipated applications worldwide in such diverse areas as
satellite imaging, computed tomography, desktop publishing,
manufacturing inspection, and atomic physics. One such application,
that of the CAT Scanner, resulted in a Nobel Prize for Sir Godfrey
Hounsfield and Alan Cormack in 1979, and that of R.Young in the
development of the scanning tunneling microscope, in 1972 at NBS,
in support of another Nobel Prize. The Nobel Prize for Physics in
1997 was awarded to William D. Phillips of NIST who used a
computer analysis of video images to determine how a cloud of atoms
spreads out as it is being laser cooled.
Thus, we see that the collective imagination of the computer community
took up the challenge raised by this new powerful tool and extended
it "far beyond our poor powers to add or detract."
Read previous mail | Read next mail
| |
