Joe Stegman's WebBlog

Hi Joe,

I cannot find the BMPDecoder source that the BMPDecoderSample references..

Thanks for the example of how to do this.  I also found an example on Ian Griffith's blog and have built some practice applications using the technique.

Based on code inspection (not profiling) you could improve the PngEncoder performance substantially by rewriting the ComputeAdler32 function.  It currently performs two modulus operations per byte endoded image (so typically six modulus operations per pixel) which is computationally expensive.  Wikipedia.org has an example of optimized adler32 code -- short explanation is to perform the modulus only every 5552 iterations, because s1 and s2 are guaranteed not to overlow a 32bit int until then.

It's unfortunate that there isn't an alternative encoding method to use for PNG.  You need to make three passes through the data -- once to copy the pixels which you do in a Write, once to computer adler32, and once to compute CRC.  Not only that, but a change to a small part of the image requires reprocessing all of it for the adler32 computation.

Michael

Apesar de andar muito ocupado nestes últimos dias, com projectos importantes e de carácter importante