I'll respond to your letters after the show... I'm very interested in what you are doing. Jeri --- "Hársfalvi, Levente" <email@example.com> wrote: > Hi! > > > I'm having some problems with the gfxconv stuff I'm working on (...as > usual > :-/ ). ...At first, sorry for the long letter; just delete if you're > not > interested in color optimization issues. > > This routine is intended to take raw 24 bit 320x400 images and > convert them > to the C64 styled, enhanced multicolor mode of Jeri's new videoboard > (in the > same resolution). I'm at the last stage of the optimizer algorithm. > Things > like reducing the number of colors, picking up a suitable value for > background color and selecting color triads for each 4x8 color blocks > are > done (and work O.K. as far as I'm concerned). > > (Just for a short explanation: the mode works exactly as the usual > multicolor bitmap mode, except for the resolution and the colors. The > resolution is just 'expanded', no real change in the organization. > For the > color thing: there is a color palette with 256 entries (instead of > the > original fixed 16 color long 'palette'). Color registers (like > background > and the others) are treated like 8 bit indexes to this palette (all 8 > bits > are used, instead of 4 bits like the original VIC did). In bitmap > modes, the > situation is similar, with one addition: the color memory values (the > usual > 4 bit nibbles) give the low 4 bit nibbles of the index. The higher 4 > bit > nibble is given by the high 4 bits of the color RAM, and this higher > 4 bit > is common for all color indexes in the respective 4x8 color block.). > > There is a problem. In MC mode, the above (last) rule means that all > 3 MC > colors for color blocks must be in one (16 colors long) palette > chunk. > (Different color blocks can select colors from different chunks, of > course). > No problem when it's just one, or a few blocks -- but this rule must > be > taken into account for _all_ such color blocks in the image, creating > a > heavy dependence between colors. > > (Quick calculation shows that if there are just 20 colors on the > whole > image, and each colors are featured with the other colors in the > image at > least once, with the above organization they occupy exactly 256 > places from > the palette.) > > I inserted a small code piece that listed out the color dependencies > on the > test image and the results are hmmm... 'embittering'. When using > color > reduction to just 64 colors for the whole image, some colors depended > on > 30-40 other colors in the map. > > I think I know the 'direction', just don't know the way. I hope > someone did > similar programming tricks, so he could give me some help. > > Imagine an X*X type symmetrical matrix, where X is the total number > of > colors in the image. If there's a '+' in the (i,j) position of the > matrix, > that means that the two colors (i, and j) were found in the same > color block > of the image at least once (ie. they're dependent). (Crosses in the > (i,i) > positions correspond to the fact that the i-th color was found in the > image > at least once (ie. 'dependent on just itself')). The problem to be > solved: > form <=16 disjunct groups from the above dependency matrix, where > each > disjunct groups include <= 16 elements, with the possible least > number of > ignored dependencies. > > ...Well, this is what is above me at the moment. Anyone with some > ideas on a > working algorithm?... > > > Thanks, > > L. > > - > This message was sent through the cbm-hackers mailing list. > To unsubscribe: echo unsubscribe | mail firstname.lastname@example.org. __________________________________________________ Do You Yahoo!? Yahoo! Auctions - buy the things you want at great prices http://auctions.yahoo.com/ - This message was sent through the cbm-hackers mailing list. To unsubscribe: echo unsubscribe | mail email@example.com.
Archive generated by hypermail 2.1.1.