mation
biggest
hat we
nts the
to gain
s in 2D
zorithm
it again
ms and
be very
Yologies
the data
most of
X/server
ansform
1g much
services
systems
ey have
nats and
y, high-
onent of
ost fields
; include
criminal
ore. The
increased
assed the
tion NOW
s film m
>xposures
expected
eras now
ie a major
ntation à
equally
| imaging
| through
w capture
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
This paper shows a new approach which uses a traditional
image visualization technology to make large ultra-high-
resolution images fast and interactive on the Web and expands
it to be used as an infrastructure of 2D applications. The
technology | is | accomplished using image streaming:
incremental, on-demand access to image data by level of detail
and area of view. It also explains how GSN’s image streaming
technology works and how it is implemented to support 2D
GSN applications: 2D viewer. It also discusses the technologies
which are used to cache huge image data set on client side.
Chapter 2 will talk about some difficulties exist in Internet-
based Image application. Two techniques pyramid and
streaming will be discussed in chapter 3. Chapter 4 presents the
implementation of the Image Streaming System — GSN image.
And a test and conclusion will be introduced in chapter 5 and
chapter 6.
2. Difficulties for Internet-GIS Image Applications
Imagery forms an essential and well-understood part of GIS
applications. There are unique challenges associated with
making effective use of imagery, especially when providing
Internet or Intranet GIS web mapping solutions. But image size
is extremely larger. Even city wide coverage with high
resolution color air photos can reach 300GB to 1.5TB for a
typical GIS web mapping application covering a city. As can be
seen, a GIS web server needs to be able to easily serve 300GB
to 40TB of imagery. Yet the benefits from using imagery are
compelling (for example, vector GIS and CAD data is often
created from imagery as a starting point), and it provides an
intuitive, accurate and up to date view of the real world for
users.
Basically said there is only one problem which needs to be
solved exists in Internet-GIS image applications: the conflict
between extremely large data set and limited Internet
bandwidth. Besides image size, network is another issue for
Internet based image applications. Internet is the most complex
network around the world. It includes a range of networks. The
increasing availability of high-bandwidth network access
greatly eases access to images of moderate quality. Images of
high quality, however, easily exceed the capacity of even the
highest volume transmission technologies. While hundreds of
kilobytes of data are commonly downloaded, megabytes, tens
of megabytes, and certainly gigabytes of data are impractical to
transfer over current infrastructure. Advanced compression
techniques achieve 4 to 1 or even greater lossless compression |
and 10 to 1 or even 100 to I lossy compression. Even in
combination with a high-bandwidth connection, however,
compression is inadequate for publishing high-quality images.
A 32 megabyte image licensed from a stock photo house — even
if compressed 100-fold with wavelet technology to
approximately 320k — is inconvenient for Web viewing. The
time required to download larger files makes them impractical
0 publish without image streaming. Only the streaming
technology can support real-time viewing of images many
gigabytes in size.
Studies have shown that around 60% of the Internet users’
access speeds are 56Kbit/s and about 30% of the Internet users
are using high speed connection. The average speed for high
Speed users are 2Mbit/s and therefore the mean connection
Speed of Internet users is IMbit/s. What will happen if we use
this bandwidth to display an image at 2048X2048. The image
Size with raw format will be 1024*1024*3*8 — 96Mbit. That
means it will take 96 second to transfer it to the user who want
to see it. One easiest way to reduce the transportation time is to
compress the image to reduce its size. For instance, we can
store it as JPEG image with ratio of 10 and the response time
will be 1/10" of the original time (9.6 second). But it is still a
problem. For normal high resolution images in JPEG format,
their size is 800Mbit, we have to figure out other solutions to
solve this problem.
3. Principle of image pyramid and image streaming
It is well known that the decompression of images will take a
lot of time and memory. Here is an example running on a
PIV2.0GHZ/500M machine. Using Adobe Photoshop6 to open
a JPEG image file will take 5 minutes and 40 seconds. It will
use 2G hard disk to store the temporal data. The 80MB file,
which covers Canada East area, is a 27845 x 23177 in width
and height and 100m resolution image. Therefore, all the
available Internet-GIS image Systems will pre-process the
images to a new format image data set first before put them to
the server. The new format is normally called image pyramid.
In order to handle the image pyramid via the Internet, a
technique which is called image streaming has been applied to
It.
As we all know, every image has a resolution. If you want to
buy a digital camera, the first question you ask yourself is what
the maximum resolution is. Different resolutions of an image
mean different quality and different size of it. In web-mapping
system, the multi-resolution approach is very useful.
o lmages usually contain features of physically significant
structure at different scales of resolution
o For some problems, this allows us to select a desired level
of detail
o For some problems, the coarse-to-fine approach can
reduce the computational complexity
© There is strong evidence that the human visual system
processes information in multi-resolution fashion
o Image compression
Pyramid representation is another important technical factor.
An image pyramid is formed by repeated smoothing and
subsampling of an image ("Subsample" — discard pixels). There
are two very common pyramid: Gaussian pyramid and
Laplacian pyramid (Burt and Adelson, 1983). Each new level
of a Gaussian pyramid is formed from the previous level by
smoothing and subsampling.
Image Streaming, on the other hand, allows the Slideshow to
Start running before all images are loaded. With Image
Streaming, an image is shown immediately once it is loaded;
while the rest of the images are being loaded in the background.
The Slideshow can even loop with this partial list of loaded
images. Eventually, all images will have completed loading and
the Slideshow will run in its full form. Image Streaming
dramatically cuts down waiting and loading time thus
improving the user's experience.
In order to implement image streaming, large images need to be
reconstructed with image pyramid and image cutting. Pyramid
is a technology where a high resolution image is created and
saved in a multi-resolution format. If you think of it as a
pyramid, with the full resolution at the base, say 3000x3000
pixels (Base layer in Figure 3.1), then 1500x1500, all the way
down to the thumbnail (Top layer in Figure 3.1), which is 10-
20K in size. If you click on the thumbnail, you are taken to the
next level of resolution. At that level, the image is broken up
1139
