I wanted to ask for some suggestion/review of a new image format I'm working on. So to answer the question: Why? The Problem: Working with a multi band scanning sensor that generates 4MP/s 16bit images (8MB/s). Sensor has 12 bands. Raw uncompressed files may run into 100's of gigs. Additionally the pixels are NOT co registered. Tiff container is out, 32bit limits are bad. BigTiff seems hackish. Jpeg2k is too slow, indexing of tiles is horrendous. pgf looks interesting, but it's got some hard coded type issues. It's also more complex than the format I'm looking at. So I'm almost 1/2 done with coding up a new format. This is also wavelet based, using a pretty new very fast algorithm. Hopefully there's not a move to patent it. For my own initial uses I'm going for keeping the container and encoding pretty separate. For now I just need single band lumanance with bilevel alpha and floating point encoding. Because the bulk of the data is bit encoded format will be MSB order with variable length addressing (address bytes as part of the header) and with unsigned values also being variable length. The format should be pretty easily mmappable and for large tiles should easily support 3 threads per tile. I'm just curious to perhaps have some folks here do a "gut check" on what I'm working on. Future implementations: Each tile will always be single band, the container will be responsible for keeping track of which tile has Red, Green or Blue. Probably will hold to a single alpha channel per container. Container will have "pluggables" to do multi band processing (YCbCr/YCgCo, etc). Probably will not allow for indexed color. Thx, Brian
Attachment:
signature.asc
Description: PGP signature
_______________________________________________ Gegl-developer mailing list Gegl-developer@xxxxxxxxxxxxxxxxxxxxxx https://lists.XCF.Berkeley.EDU/mailman/listinfo/gegl-developer
