raster - Wavelet vs Discrete compression for imagery

What is the main reasons to choose wavelet (jp2, sid, ecw) instead of discrete (JPEG) compression methods for raster data?

Lets limit this question to aerial imagery and lossy compression.

Why should we give preference to wavelet compression instead of "classic" JPEG? It will give smaller size but may be there any drawbacks, what do you think? Will it lead to more CPU processing for decomression (during viewing)?

Answer

I suspect that there are a great number of factors that go into the choice of image format and compression scheme:

• Image dimensions


• Bit depth


• Image complexity (images with large areas of similar colors may actually compress better by a lossless codec than a lossy codec, and some codecs handle complex, detailed areas better than others)


• Multi-band support (e.g. 4-band TIFF which allows both true color and color infrared products to be produced from the same image)


• Alpha channel/transparency support


• Compression ratio


• Compression speed and resource requirements



• Decompression speed and resource requirements


• Compatibility (what technologies, hardware and software are involved?)


• Use case/analysis requirements (Is it going to be served over the web or used in a desktop GIS? Is the imagery going to be used only for visual reference, e.g. as a basemap, or will it be used in analyses where the accuracy of individual pixel measurements is important?)

The larger the file (before compression) the more benefit you are likely to get from wavelet compression, lending themselves to ECW, JPEG2000, etc., while with tiled image services like Google Maps which are made up of many small files, you want them to transmit them over the web and decompress the entire file in a web browser quickly so JPEG or PNG make good sense.

Additionally some wavelet algorithms like MrSID allow for quickly viewing/extracting a subset of a large image by decompressing just the area of interest rather the entire file, making them good for large, high-resolution images that you only look at small portions of most of the time.

The bottom line is that different requirements and data sources lend themselves better to different data compression algorithms, which is probably one good reason as to why there are so many, and why there is no one-size-fits-all solution.

Further reading: