TEACH CONVOLVE                               Allan Ramsay, November 1983

CONVOLVE provides a  demonstration of  some of the  techniques used  for
processing grey-level  pictures.  It contains  routines  for  displaying
grey-level pictures,  running masks  over them  to find  edges or  bars,
removing extra edges where a gradual intensity change leads a given mask
to respond more than once, and thresholding the set of responses so that
only significant ones are retained. It is also possible to run an  image
right through  all these  processes, with  the intermediate  "convolved"
picture being displayed at every stage.

The following procedures are provided:

NORMALISE(PICTURE); takes a turtle picture which has arbitrary  integers
as the values of its pixels, and returns a similar picture where all the
values are between 0 and 9.

GREY_DISPLAY(PICTURE); takes a normalised picture, replaces each integer
value by a  character whose brightness  (on the screen)  is supposed  to
correspond to  the  value of  the  integer,  and displays  this  on  the
terminal. Thus a pixel with  value 0 is displayed  as a space, one  with
value 1 is displayed as a "." ...  and one with value 9 is displayed  as
"#". This is clearly not ideal, but it is surprisingly effective.

CONVOLVE(PICTURE, MASK); runs  the mask over  the (grey-level)  picture,
and produces a new picture with the  response of the mask at each  pixel
used as the new value at that point. The mask should be a list of  lists
of values, with each list of  values corresponding to the weights to  be
used on the points in a given column (i.e. the list [[1 0 -1]] describes
a vertical  mask, where  the value  for the  pixel above  the one  we're
considering is to be multiplied by 1,  the value of the pixel itself  by
0, and  the  value of  the  one below  by  -1, whereas  [[1]  [0]  [-1]]
describes a horizontal mask with the value  of pixel to the left of  the
current one being multiplied by 1 and the value of the one on the  right
being multiplied by -1).

FINDPEAKS(PICTURE,MASK); produces a new  picture with the local  maximum
values of the given picture marked, and the others zeroed out. The  mask
is used to define  "local" - FINDPEAKS scans  the region covered by  the
mask to see if it contains a  value greater than the one at the  current
pixel, and only copies  the current value across  if it it doesn't  find
one.

THRESHOLD(PICTURE,VALUE); produces a new  picture with spaces where  the
old one had  integer values  less than VALUE,  *'s where  it had  values
greater than or equal to it. This  is to help you see what lines,  bars,
etc. you have found by the previous processes.

PROCESSPICTURE(PICTURE,MASK,PEAKMASK); runs  CONVOLVE over  the  picture
with the given  mask, runs FINDPEAKS  over the result  of this with  the
given peak mask, and finally runs THRESHOLD over the result of this with
value 2.

EXAMPLE(PICTURE); runs  all  three  procedures over  the  given  picture
twice, once  with a  basic mask  and a  peak mask  for finding  vertical
edges, once with a pair of  masks for find horizontal ones, and  finally
produces a picture with all the edges found by either filled in.

All  these  procedures  provide  displays  of  their  input  and  output
pictures, so long as the variable CHATTY is set to TRUE. Its default  it
FALSE, but inside EXAMPLE it is set to TRUE.

There are examples of grey level pictures stored in the following files:

    LIB GREYPIC
    LIB PERSONPIC

When you  load one  of these  files,  the picture  it contains  will  be
assigned to the  global variable PICTURE.  So if you  want to have  more
than one of them loaded at once, it is advisable to type e.g.

    lib greypic; picture -> greypicture; ;;;(DON'T store it in greypic, since
                                         ;;; this already has a value !!!)
    lib personpic; picture -> personpic;

You can then have any of these pictures analysed by e.g.

    example(personpic);

and you can store the resulting  picture containing the edges found  for
this picture by

    picture -> personedges;

--- C.all/teach/convolve -----------------------------------------------
--- Copyright University of Sussex 1988. All rights reserved. ----------