People‎ > ‎John Stavrakakis‎ > ‎John's blog‎ > ‎

Improved compression performance

posted 1 Dec 2011, 02:26 by John Stavrakakis
I've been toying with some ideas about how to increase the compression performance of the bit plane images before passing it over to the general compressor software zlib. Using several different approaches, I have found these redundancies in a deterministic manner, but I intend to improve these with a statistical approach. There are 3 methods, The improvements to raw data that needs to be compressed: 

Bunny model
Method 1 (index based):             46.8%
Method 2 (sequence based):       5.1%
Method 3 (double indexed):         7.2%

Randomised data
Method 1 (index based):             23.9%
Method 2 (sequence based):       8.2%
Method 3 (double indexed):         19.7%

I still need more time to work on method 3. Its quite interesting because I haven't fully taken advantage of the method. I effectively reduce the number of 1 bits in the image by 3! by only using a single number. Whats more, the index is based on local position, so this number is bounded by the amount of sparseness, thus being log_2(local_data_length). The good thing is that each of these methods could potentially be performed in sequence, the known savings for each in the compressed case is not known yet. For method 1, the compressed improvement is 
16.4%. The reported raw data reductions for the randomised data is quite necessary since for any data set with n points there will be an expected gain. This is good for more effective memory allocation, but also the robustness of the method for any given input, just specify resolution, and number of visible points. The randomised data for raw reduction has very interesting properties, while I am eager to explore, this is actually work that I should not necessarily be focused on at the moment grrrr 

Ah the usual Xcode crashy crashy, but somehow I am cursed because MS visual studio 2010 on another machine is crashing lately. I may have also corrupted my gfx card. When the Direct X call for Present is used, the clear colours are a bizarre alternation of bright green and pink...will Nvidia/ATI hire me to test their new hardware and windows drivers? I seem to be isolated with such phenomena