On Tue, Jan 11, 2011 at 1:59 PM, Anand Balachandran Pillai < abpil...@gmail.com> wrote:
> > Correction - "as moderator of this forum", still you can go and take > a hike. > :) Anywayz, i was planning to post the following reply initially when this thread started, but opined that i will first understand how OP is thinking : Many use thresholds to convert images - esp when doing a color-b/w. Though this works most of the time, it is not recommended; there are standard image processing algorithms which use histograms or kmeans to accomplish this. Halftoning is an interesting problem and similar to different algorithms that exist for text segmentation/classification, this is a 'classic' one. Blindly using PIL for images is not a good idea. Image processing algorithms are *extremely* expensive, as they process huge quantum of information - hence you have DIP processors, though there is nothing equivalent like a 'text processor' (IMHO). DIP libraries provide loads of functions for developers to use; they abstract the complexities of the underlying algorithm, and hence i think the responsibility on the developer increases more now, as he/she needs to really understand what he wants to accomplish with the pixel data. Iterating the pixels multiple times for simple transformations is not a good idea; especially when the code is to be deployed in production servers. I used to actively work in DIP during my undergrad years(robotic vision), but offlate hardly touch this area. Just hanging around Stackoverflow <http://stackoverflow.com/questions/tagged/image-processing>to keep a watch on the Q&As etc is a great way to learn too; otherwise, Gonzalez and Woods is an amazing book to start; especially the edition that came out in 2004 had a chapter on wavelets which was fun. -V http://blizzardzblogs.blogspot.com/ _______________________________________________ BangPypers mailing list BangPypers@python.org http://mail.python.org/mailman/listinfo/bangpypers
