Friday, October 31, 2014

[ftsprcam] Compressing floating point images

Round the exponent up to the nearest multiple of 8 and shift the mantissa to the left with zeros to keep the value identical (making it up to 8 bits wider).  Compress the sign bit, exponent, and mantissa separately using standard lossless integer-sampled image compression, e.g., JPEG 2000 Lossless.  The mantissa probably needs to get sliced bitwise into several 8-bit planes.

Motivation was to compress intermediate computations of the Mandelbrot set with real (continuous) escape times.

No comments :