00100	SAILON NUMBER 71.		                    DRAFT CRE MANUAL.
00200	
00300	
00400	STANFORD ARTIFICIAL INTELLIGENCE LABORATORY	        JANUARY 1973.
00500	OPERATING NOTE NUMBER 71.
00600	
00700	
00800	
00900	                          - - - DRAFT - - -
01000	              Contour-Region-Edge Image Representation.
01100	
01200	
01300	                          Bruce g. Baumgart
01400	
01500	
01600	ABSTRACT:
01700	
01800		A  contour-region-edge,  CRE,  image representation is stated
01900	and an algorithm for converting a digital television image into  this
02000	format  is  explained. An implementation of the algorithm is the main
02100	routine of a program  called  Cart's  Eye,  CRE;  auxiliary  routines
02200	provide  manual  cart  control,  TV  camera input, image display, and
02300	Xerox Graphics Printer output. Potential application of CRE as a step
02400	in object perception is discussed.
02500	
02600	
02700		CONTENTS:
02800	
02900			I.
03000			    A. 	DATA STRUCTURE.
03100			    B.	THE ALGORITHM.
03200			    C.  PERFORMANCE.
03300		 	II.
03400			    A.	TELETYPE COMMANDS.
03500			    B.	FILE FORMATS.
03600			    C.	SAIL INTERFACING.
03700			    D.	LISP INTERFACING.
03800			III.
03900			    A.	EDITORIAL REMARKS.
04000			    B.	CODE DOCUMENTATION.
04100			    C.	APPENDICES.
04200		
04300	
04400	This  research was supported by the Advanced Research Projects Agency
04500	   of the Office of the Secretary of Defense under contract SD-183.
     

00100	INTRODUCTION.
00200	
00300	
00400		CRE  is  a  solution  to  the  problem  of  finding  all  the
00500	photometric  edges in a television picture and the regions bounded by
00600	those edges; this is done by building the edges and regions on top of
00700	a  very literal intensity contour map. Edges are formed where contour
00800	lines run close together and regions are closed by following  contour
00900	lines  across "flatlands" until they return to their edges.  Also CRE
01000	links regions and edges of one image with corresponding  regions  and
01100	edges of the previous image.
01200	
01300		The process is automatic and is intended to run without human
01400	intervention.   Furthermore, the process is "bottom up"; there are no
01500	significant  inputs  other  than the given television images. The CRE
01600	design goal is to provide video image input  data  in  a  basic  list
01700	structured  form  appropriate  for my geometric editor, GEOMED; other
01800	aspects  of  the  vision  problem  such  as  3D  perception,   object
01900	recognition, wide angle comparison, photometric normalization, and so
02000	on are not attempted in CRE.