A 'Hyper-Pyramid' Architecture for Massively Parallel Image Processing

The rmage processing involves many compuial~on on a Eargt data volume. I V c shorn that the "dtczde and canqucr" mrihod may br uacd 80 decreose t h r CEPcution hmc of a lgonfhms. LVe need la hove a suttable topology l o profit jully of dalo pnml!elrsrn. The lapologrca oa mcah or pyrnrntd arc loo much r e s t m ~ ntng on the parsrh~ltt~cs af allowed m o r e m ~ n t dala . I n other w a y the h y p c r c t l b e fopology has a Iorge tctnng rornplexr ty . Al.ra we proposr a hgper-pymnttd lopology , tuhach ts a rompromrve beiwecrz wrnr ig cnrnplertly and a lgor t thrn tc carnp l~rhy . Thus the h~.~togrum and component l abr l tng olgonlhmn o n resprr t tv~lg an olgonthrnlc compI~r~iy jtt Q( Vlog n) and rn O( Iog271). Whew Y :a the number o j colour rn Image 01 n p ~ r ~ l s .