perm filename 2[CRE,BGB] blob
sn#021791 filedate 1973-01-25 generic text, type T, neo UTF8
00100 THE ALGORITHM.
00200
00300 In the large, CRE consists of five steps: thresholding,
00400 contouring, smoothing, bundling and comparing. The first four steps
00500 perform conversions between five kinds of images: 6-bit raster
00600 image, 1-bit raster image, vector intensity contour image, arc
00700 contour image, and winged edge image. The final step, comparing,
00800 joins an image to a film of images by linking the parts of the
00900 cuurent image to the parts of the previous image that compare nearly
01000 equal.
01100
01200 MAJOR OPERATION. OPERAND. RESULT.
01300
01400 1. THRESHOLDING: 6-BIT-IMAGE 1-BIT-IMAGES.
01500 2. CONTOURING: 1-BIT-IMAGES VIC-IMAGE.
01600 3. SMOOTHING: VIC-IMAGE ARC-IMAGE.
01700 4. BUNDLING: ARC-IMAGE WINGED-IMAGE.
01800 5. COMPARING: IMAGE & FILM FILM.
01900
00100 THRESHOLDING.
00200
00300 Thresholding, the first and easiest step, consists of two subroutines:
00400
00500 1. THRESH: CUT,TVBUF → PAC
00600 2. PACXOR: PAC → HSEG,VSEG
00700
00800 The subroutine THRESH takes an integer argument, 0 < CUT ≤ 63, and
00900 for each pixel in the TVBUF with value equal to or greater than the
01000 cut THRESH sets a bit in PAC. PAC (picture accumulator) is a bit
01100 array of 216 rows by 288 columns which takes 1728 words in the TVSEG.
01200
01300 The subroutine PACXOR first copies the PAC into two slightly larger
01400 bit arrays named HSEG and VSEG, then it exclusive OR's the PAC,
01500 properly displaced one row or one column, into HSEG and VSEG to
01600 compute the horizontal and vertical border bits of blobs in the PAC.
01700
01800
00100 CONTOURING.
00200
00300 Contouring, the next major step, converts the bit arrays HSEG
00400 and VSEG into a node-link data structure that represents an equal
00500 intensity level contour map. Of such contouring, there be two minor
00600 steps: one, that of tracing the contour as a ring of vectors to form
00700 a polygon; and two, that of placing the polygon in the contour tree
00800 data structure.
00900
01000 Although the notion of a contour
01100 map is familiar to everyone as a piece of paper from the Geodetic
01200 Survey Office; implementing the notion into a data structure becomes
01300 a magical mystery tour. Two of the contouring mysteries to be
01400 discussed are jaggies and nesting. The jaggies problem involves doing
01500 something to a rectilinear quantized set of segments, to make its
01600 linear nature more evident. The jaggies solution in CRE is called
01700 the dekinking, and merely involves vernier positioning the
01800 right-turns as will be explained below.
01900
02000 A JAGGY ILLUSTRATED.
02100
02200 ___
02300 |___
02400 |____
02500 |___
02600 |___
02700
02800 The nesting problem involves comparing two polygons and deciding
02900 whether one is within the other; although easy in an isolated case,
03000 solving alot of nesting becomes very expensive in compute time or in
03100 memory space. The nesting solution in CRE is a fast big memory one
03200 involving a 62K array, called the SKYSEG.
03300
03400
03500 SMOOTHING.
03600
03700 BUNDLING.