COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	He  proposes  to  extend previous  work  by  varying  parameters  and
C00005 ENDMK
C⊗;
He  proposes  to  extend previous  work  by  varying  parameters  and
thresholds,  by   utilizing  shadow,  highlight  and  multiple  image
information, and by using certain topological constraints.  I support
the extension  of this work.   I have some questions  directed to the
science  content of the work.   I think that  the field is important;
the  approach of  applying  a  local edge  description  operator  and
linking local edge  fragments is significant.  Others  are at work on
the problem.
  1.   The setting of  parameters and  thresholds will be  empirical.
  How  can the scenes  and operators be  modelled so  as to calculate
  thresholds?  
  2.  What is  the computational complexity of graph-searching?   How
  does it compare to other possible methods?  How does it behave with
  increasing resolution?  
  3.  A  mention is made of  applicability to textured images.   What
  explicitly  is   meant?    There  are   strong  limits  imposed  by
  resolution.  Certain boundaries in textured images are  not defined
  by any "strokes".  Clarify.  
  4. Not much use is to be made of multiple images.  
  5.  Vague about  how to distinguish shadows and  highlights and how
  to use them. 
There are some technical points directed at the exposition. 
  a.  What use is  made of the assumption that shadows  are cast only
  by objects in the scene?  
  b.  p3 "exchanged"--what does it mean?  
  c.   p4 There is  no setting  of threshold so  as to be  reasonably
  certain ......  The contrast across edges can be arbitrarily small.