Choosing the most appropriate raster format for your project

There are many raster formats to choice from for your mobile GIS project, each with varying image properties and subsequent performance impact. Some understanding of image properties is required in order to be able to choose the most appropriate raster for your work.

Images are sometimes referred to by a combination of the number of bits and the number of samples (or 'bands' for TIFF images). For example, an 8 bit TIFF with 3 bands may be referred to as a 24 bit TIFF. The number of colours that are available in any given raster is a function of the number of bits and the number of samples. For example, our 8 bit TIFF with 3 bands will have 16 million colours (i.e.: 23x8 colours).

The following table shows a comparison of raster formats (and their characteristics) that are supported by ArcPad.

Format

Compression

Number of bits

Number of samples

(or bands)

Number of colours

Lossy?

TIFF/uncompressed

1:1

1

4

8

16

1

1

1 or 3

1 or 3

2

16

256 or 16 million

65536 or 7.9 x 1028

no

TIFF/LZW

10:1

1

4

8

16

1

1

1 or 3

1 or 3

2

16

256 or 16 million

65536 or 7.9 x 1028

no

TIFF/Packbits

2:1

1

4

8

16

1

1

1 or 3

1 or 3

2

16

256 or 16 million

65536 or 7.9 x 1028

no

TIFF/CCITT

100:1

2

1

2

no

TIFF/Deflate

10:1

1

4

8

16

1

1

1 or 3

1 or 3

2

16

256 or 16 million

65536 or 7.9 x 1028

no

CADRG

55:1

8

1

256

no

GIF/LZW

10:1

8

1

256

no

PNG

10:1

1

2

4

8

16

1

1

1

1 or 3

1 or 3

2

4

16

256 or 16 million

65536 or 7.9 x 1028

no

JPEG

20:1

8

3

16 million

yes

JPEG2000

20:1

8

16

1 or 3

1 or 3

256 or 16 million

65536 or 7.9 x 1028

yes

MrSID

20:1

8

16

1 or 3

1 or 3

256 or 16 million

65536 or 7.9 x 1028

yes

ECW*

20:1

8

3

16 million

yes

The following table summarizes the description of memory footprint and drawing behaviour for each of these formats. These descriptions are subjective and have been deduced from our experiences with these formats.

Format

Memory footprint

Drawing Behaviour

TIFF/uncompressed

tiny

extremely fast because it loads the minimum rows to draw

TIFF/LZW

tiny

fast because it loads the minimum rows/tiles to draw

TIFF/Packbits

tiny

fast because it loads the minimum rows/tiles to draw

TIFF/CCITT

tiny

fast because it loads the minimum rows/tiles to draw

TIFF/DEFLATE

tiny

fast because it loads the minimum rows/tiles to draw

CADRG

tiny

fast because CADRG files are usually small files

GIF/LZW

tiny

very slow because decompresses the whole file

PNG

tiny

slow because decompresses the whole file

JPEG

large

extremely fast, because an uncompressed copy of the image is used for drawing

JPEG2000

medium

slow because the decompression is CPU intensive

MrSID

medium

slow because the decompression is CPU intensive

ECW*

medium

slow because the decompression is CPU intensive

* ECW requires a 3rd party extension, ER Mapper's "ArcPad v6.0 ECW Plugin 1.2" and may not support all features of ArcPad, such as map rotation.

Below is some explanation of these raster formats that may help you determine which is best for you.

TIFF is generally regarded to be the best format for ArcPad. It draws fast, can compress reasonably small, and the decompression doesn't consume significant CPU power nor use significant amounts of main memory. ArcPad can make use of information in the TIFF header to minimally retrieve only the rows/tiles of data that are needed to draw the image.

TIFF/uncompressed is extremely fast to draw because ArcPad knows where to look in the file to retrieve exactly the right data to draw the image. There's little demand on CPU power because there is no need for a decompression. This format is suitable in terms of speed for both small and large images alike. Its only disadvantage (although becoming less of an issue with modern devices) is the file size and therefore it's ability to be placed on a compact flash or storage card.

TIFF/LZW and TIFF/Deflate are both good. They both achieve decent 10:1 compression, require small file sizes, use little memory to decompress and are relatively fast. TIFF/LZW is widely supported whereas TIFF/DEFLATE (sometimes called ZIP-in-TIFF) is supported mainly by Adobe software.

TIFF/Packbits is generally not recommended. The scheme works extremely well when the image contains very simple shapes. Images that have complex shapes (ie: maps) do not compress well.

TIFF/CCITT is tailored for black/white fax transmissions. It assumes that a large portion of your document is white and assigns a highly compressed binary code for such an area. Such a technique can yield up to 100:1 compression ratios, however, in practice, 50:1 may be more realistic.

GIF and PNG are widely used on the internet and both achieve a decent compression of 10:1. These formats are ideal for small images, but are definitely not ideal for large images since both of these formats are sequential, meaning they must be read from start to finish. For example, if you zoom in to the centre of one of these images, ArcPad cannot jump to the right place in the file, it must always read from the beginning of the file and often has to decode large volumes of off-screen data, before it finds the portion required.

CADRG file format is used by the National Geospatial Intelligence Agency and is claimed to have compression ratios up to 55:1. However it appears that the 55:1 may be an exaggeration, due to mixing of 8 bit and 24 bit file formats, and the real figure lies between 10:1 and 20:1. The file sizes are usually small. CADRG files do not use much memory and are fast to draw in ArcPad. The only limitation is that the format is extremely restrictive in its geographic area and resolution.

JPEG, JPEG2000, MrSID and ECW are all lossy formats. They achieve high 20:1 compression by losing/replacing bits of image often undetected by the human eye. These formats are best suited for photo realistic images such as scanned images, digital camera photos and aerial photography. These formats are not suitable for computer generated drawings such as cartographic maps, because lossy compression will noticeably degrade sharp edges of lines, solid fills of polygons, and blur point data. MrSID, JPEG2000 and ECW all aim to reduce storage card memory at the expense of CPU processing power, and sometimes main memory. These formats were especially popular when compact flash (CF) and storage digital (SD) media had small capacities (eg: 64MB) and were very expensive. The 20:1 compression ratio made it possible to use images as basemaps in your mobile GIS. Today, because CF and SD media capacities have grown dramatically (eg: 8GB) and are inexpensive, the need to save so much space has become less important. These formats are no longer attractive for their 20:1 compression ratio, and have become unattractive because of their demands on CPU processing power, slowness, degradation of image quality and their sometimes high consumption of main memory.


2/6/2013