当前位置: 代码迷 >> 共享软件 >> :翻译下文,能翻多少是多少
  详细解决方案

:翻译下文,能翻多少是多少

热度:8508   发布时间:2013-02-26 00:00:00.0
求助:翻译下文,能翻多少是多少
8.   Advance   the   top   and   bottom   cutting   lines   down   or   up,   respectively,   provided   that   the   difference   is   less   than   a   threshold   θ3   and   the   luminosity   of   the   current   cutting   line   does   not   significantly   differ   from   that   of   its   predecessor.   (We   are   eating   up   a   marginal   artifact.)
9.   Write   out   the   portion   of   the   image   that   is   between   the   four   cutting   lines   to   a   file.
4.   Multiresolution   segmentation   method  
Our   method   of   segmenting   and   classifying   regions   of   a   page   has   been   described   previously   in   [15].   By   using   a   multiresolution   approach,   we   gain   two   benefits:   (a)   the   essential   image   information   for   a   structural   analysis   of   the   page   is   treated   at   a   level   of   resolution   in   which   high-frequency   noise   has   been   filtered   out   thereby   preventing   the   introduction   of   small   noise   regions,   and   (b)   speed   is   gained   by   performing   most   of   the   steps   of   analysis   on   reduced   resolution   representation   of   the   image.  
The   algorithm   proceeds   in   two   phases.   During   the   first   phase,   four   feature   pyramids   are   constructed   from   the   document   image.   During   the   second   phase,   regions   of   the   image   are   classified   according   to   rules   into   categories:   background,   text,   line-drawing,   photograph,   or   unknown.  
4.1.   Phase   1:   construction   of   feature   pyramids  
The   first   phase   constructs   four   pyramids.   First   four   “feature   images”   are   computed   from   the   original:   a   “mean”   image,   a   median   image,   a   variance   image,   and   a   “threshold”   image.   Let   the   width   and   height   of   the   original   image   be   m   and   n,   respectively.   The   mean   image   of   size     m/16     by     n/16     is   obtained   by   computing   the   average   pixel   value   for   each   16×16   block   of   pixels   in   the   original.   The   three   other   feature   maps   have   the   same   dimensions   as   the   mean   image,   but   they   contain   the   median,   variance,   and   “threshold”   values   over   the   16×16   blocks   rather   than   the   mean   value.   The   threshold   image   is   obtained   by   counting   the   number   of   pixels   within   each   16×16   block   that   exceed   a   threshold   θ4=250   in   the   original   image.  
  相关解决方案