July 28, 2005 - Image Server was announced at the Plenary Session on Monday, and yet its powerful functionality was only touched upon at that time. Image Server serves up imagery and processes imagery in a server environment on request from the client. It can link the acquisition of images through the server very quickly and serve it up in real time. With Image Server you can add additional services to the server dynamically. Image Server has access to raw files and can display imagery from different viewpoints.
Key functionality of ESRI's Image Server includes
- Fast access to large image datasets
- On-the-fly server-based image processing
- Production of multiple imagery products from a single source
- Multiplatform GIS/CAD/Web client access
- Direct access to multiple file formats and compression
- Fully scalable enterprise client/server architecture
- Data security and access logging
- Independence from third-party software or DBMS
- Expandability through extensions and SDK
- Rapid access to data without preprocessing
- No DBMS load
- Very fast access to large quantities of images
- On-the-fly server side processing can access from a small device, as well as workstations with a thin client
- Derive multiple products from single source with different process attached to the data
- Dynamic update of image services
A demonstration given by Peter Becker, clearly showed the power of this product as it was running on the desktop with 2 gigabytes of RAM. The example in this case used Landsat imagery, a dataset freely available from 8,000 Landsat scenes, with tiff files with 8 1/2 Terabytes of managed overview. If anything changes underneath it, the overview will change.
Image Server lets you zoom and pan around the dataset. The server is very busy: it's back looking through an index and extracting pixels from various scenes, it's doing coordinate transformation, it's being fed a mosaic and then displaying on the screen. As you zoom in it's actually adding pan sharpening to the process. Most bands are 50 meter, now we have a pan sharpened image. You can change band combinations, compute them, for example, to a false color infrared. The server goes back and done computations. NDVI is essentially just algebra being applied to the bands.
"We can handle SRTM data, we talk about imagery, not just standard pixels, or colors. They can actually be terrain modeled. 40,000 tiles - the server on-the-fly identifies what areas I need."
Another dataset: scanned maps 1:500,000 Russian maps of the world. What is actually stored on the server are the scanned maps that came in as tiff files put on the server. Someone has measured the tick marks and georeferenced them. "From those parameters, every time I zoom it clips the images, coordinates them, mosaicking the images together and displaying them on the screen. If I change the tick marks the image will dynamically update."
It is a known fact that any time you do anything with image processing, you lose information. Every time you subsample an image, you lose. With Image Server, the image goes to the server, passes pixels through the system, subsampling is passed through with minimal if no degradation.
Another example: A service is showing Quickbird imagery. You can see the image quality, down to the struts on sides of the wings on a plane in the Dubai airport. Stored here are 16 bit basic images of unprocessed data. The process chain is as follows: take radiometric enhancement, then orthorectification, the terrain model is SRTM, then pan sharpening (to color band and pan band and merging this together), then the result is a 60 centimeter color image generated from a raw image.
"I can take pan bands raw pan imagery, can change it to get 4:1 false color, if I want to do a vegetation analysis, the quality of pixels is amazing. I'm doing an NDVI, combining that with water detection algorithms. I can add algorithms without adding data volume," explained Becker.
"You can change properties of imagery, for example, I can use two services -- one with JPEG 15 and one without. I can add an additional process to the process change--I'm compressing it after I've done the processing. I can pan and zoom around my dataset even on a low band connection and find what I'm looking for." On PDAs, where you don't have the bandwidth and image quality you would like, this is important.
Becker said that you can also get metadata. You can change properties, can change super sampling, and get better information about the service you're displaying as well as the meta information about each of the various images. You can see in XML the bands that have been used, the radiometry that has been used, and these have been transmitted by the server application. Meta information can be controlled by the administrator. "You might be dealing with secret information and not want to share it," he pointed out.
In a digital aerial photography example, the photography had been done by a digital ortho camera, with images that are processed on-the-fly by the server. The digital camera data has been loaded onto the server, and has also orthorectification parameters, going into the database. "We have an existing terrain model. We have on-the-fly orthophotos generated by the system. That imagery is available very quickly and get the product right away," said Becker. "The accuracy may not be what you want right away but for many applications that is important, rather than waiting months for a product."
What is the importance of imagery for GIS applications?
- Natural background - for many GIS apps
- Analytical dataset - statistics (E.g. NDVI)
- terrain models used often for background for doing data sets, where does vector data come from? Someone has to extract it
- Source of most vector map data - extraction (vector mapping) you can get imagery quickly and cheaper than vector data.
- Do we trust vector data? Imagery is used to give trust - yes it is used to verify
- Verification of vector data
- Satellites (Landsat, Quickbird, modis spot)
- Digital cameras
- Scanned aerial photographs
- Orthoimagery (DOQQ)
- Terrain models
- Scanned existing maps
- Scanned cadastre
- Scanned as-built drawings
- Archives of airphotos
- Volumes growing exponentially (30%/year)
- Increasingly inexpensive
- Plenty available
- Only fraction is accessed
- Problem is accessibility
- Volume - magnitudes larger than vector data sets
- Value - devalues quickly with time, increases again with time - steep drop in value of imagery - goes down quickly as new imagery is taken, reduce time it takes between gathering of imagery and use of it.
- Fixed - primary data does not change, output from processing changes-it's a snapshot of the world in one instant of time.
- Determine parameters
Other Important Features
Image Server has direct access to native file formats -
Geotiff, JPEG, LZW, PackBits, flat, scanline, tiled
BIL, BIP, BSQ
JPG2000, MrSID, ECW
SDK to add additional formats
- Store and process uncompressed or compressed imagery
- Compress for transmission recommend not compressing data (raw data)
- diskspace is cheap
- higher quality
- create more products
- Can applied to each individual image
- can be different for each image
- process parameters
- in process definition (XML)
- fields in the database
- sample input processes
- extract bands and band algebra
- sub and super sample
- Service has default projection
- Client can specify output projection
- Fuse high resolution pan with lower resolution color
- Very fast
- Very accurate
- Fusing of multiple input images
- Important to be able to control how images are put together
- Can see same viewpoint from different views
You can create a service called My World, where you're able to create a very high resolution terrain model of the whole world, which is important for different types of terrain analysis. You can then use low resolution data where you don't have good data, and then higher where you have a higher level of interest.
Besides the processes already in Image Server you can
- Add your own process
- Specialized processing
- Proprietary algorithms
- Interface to hierarchical data storage
- Interface to payment gateway.
- Load balancing between multiple servers Can have same service on multiple servers and servers will perform load balancing between them
- Recommend hard disks stored with the server
- Cascade servers - meaning, have servers requests service from another service, while the end users see it as the world
The Server creates overviews of images as you don't want the server displaying all the images it has. These overviews create 1% to 30% additional space so the display is very fast. With tiles the server can monitor every single request made, and there is a log to see who is requesting what data. 90% of requests are made in 5% of the area. Along corridors of new development, towns, and other areas you have random requests.
You can define areas that are high priority and the server will preprocess those areas and store them by tiles. If someone zooms into that area, it's already processed. This way you can reduce the amount of load on the system.
Image Server is a client server architecture, and the client has no access to system. There is no need to convert data, as you can implement on whatever you have at the moment. Works on legacy applications, and you can still work from the same imagery, and can use any imagery for other uses.
Image Server Publisher is another product, designed for allowing organizations to have a server, copy the server and provide it to another client. This way you can deliver data to different people or can use it in server farms
Image Server and ArcSDE are complementary.
What's the difference between ArcSDE and Image Server?
ArcSDE uses database management system to store need DBMS, and database administrator. For security, a DBMS is a good choice, as you can store all vector, raster, etc. in one location, Complete products image processing takes place on client, and integration with vectors is on a common server.
With Image Server, data is stored in files, you don't have to preprocess, you have high latency, processing takes place on a server, you can work on thin clients, with integration with vectors on a separate server and source. The long term goal is to combine the two so you can work on the server or client side.
The Image Server products include:
- Image Server
- Service Manager
- ArcGIS, ArcEngine and server plugins
- ERDAS Imagine
- Intergraph GeoMedia
- MapInfo Pro
- AutoCAD and MicroStation
- Open standards such as WMS and HTML viewers