convolution kernal examples

Convolution kernal examples

Convolution kernals are templates used to describe the effect of a region of pixels on a single pixel. The template is usually located over the pixel in question such that the pixel is in center of the kernal. When using 'pip-fixed-kernal' the kernal is applied to all the pixels of the image, in turn. Specific subordinate regions of an image may be addressed using 'pip-pixel-kernal', as shown in Example 3.

Example 1 :
This example uses a very simple 3x3 edge detection convolution kernal. Edge detection kernals are recognized by the sum of the elements of the kernal is zero (0.0).

ox--> (define ratimg (pip-formatted-read 'filename "Rat_w_cov.tiff"))

ox--> (define edgelist (list -1.0 -1.0 0.0
                             -1.0  0.0 1.0
                              0.0  1.0 1.0))
ox--> (define edgevec (list->vector edgelist))


ox--> ;; Process red, green and blue fields 
ox--> (define redch (pip-fixed-kernal (car ratimg) edgevec))
ox--> (define grnch (pip-fixed-kernal (cadr ratimg) edgevec))
ox--> (define bluch (pip-fixed-kernal (caddr ratimg) edgevec))

ox--> ;; Edge image saved as an image of the red field.
ox--> ;; This gives a nice grey scale image.  Otherwise, the colors
ox--> ;; separate when applying the kernal to each field and combining
ox--> ;; the fields back together.
ox--> (pip-formatted-save 'imglist (list redch) 'filename "Edge.tiff")

Edges of Bopper Image

source image	image of applied edge kernal

Example 2 :
This example is one of blurring convolution kernals. One characteristic of a blurring kernal is the sum of the elements of the kernal is one (1.0).

ox--> (define blurlist (list 0.04 0.04 0.04 0.04 0.04
                             0.04 0.04 0.04 0.04 0.04
			     0.04 0.04 0.04 0.04 0.04
			     0.04 0.04 0.04 0.04 0.04
			     0.04 0.04 0.04 0.04 0.04))

ox--> (define blurvec (list->vector blurlist))

ox--> (define lilimg (pip-formatted-read 'filename "Lilies_w_cov.tiff"))

ox--> (set! redch (pip-fixed-kernal (car lilimg) blurvec))
ox--> (set! grnch (pip-fixed-kernal (cadr lilimg) blurvec))
ox--> (set! bluch (pip-fixed-kernal (caddr lilimg) blurvec))
ox--> (pip-formatted-save 'imglist (list redch grnch bluch (cadddr ratimg))
                          'ioform  "JPEG"
                          'filename "Blurred.jpeg")

Blurred Pond

source image	image of applied blur kernal

Example 3, using "pip-pixel-kernal" :
An 11x11 blur convolution kernal is used to blur just the region about the head of the charater in the image in this example. A transitional copy of the image is created to preserve the content of the rest of the image.

ox--> (define cellvalue (/ 1.0 121.0))
ox--> (define blurvec3 (make-vector 121 cellvalue))

ox--> (define ratimg (pip-formatted-read 'filename "Rat.tiff"))
ox--> (define redch (pip-imgfld-copy (car ratimg)))
ox--> (define grnch (pip-imgfld-copy (cadr ratimg)))
ox--> (define bluch (pip-imgfld-copy (caddr ratimg)))

ox--> ;; Define a working transitional copy of the image in 
ox--> ;; question.      
ox--> (define Rred (pip-imgfld-copy (car ratimg)))
ox--> (define Rgrn (pip-imgfld-copy (cadr ratimg)))
ox--> (define Rblu (pip-imgfld-copy (caddr ratimg)))

ox--> (define xdist 0.0)
ox--> (define ydist 0.0)
ox--> (define dist 0.0)

ox--> (for i 320 440 1
ox-->      (for j 90 210 1
ox-->           (set! xdist (exact->inexact (- 380 i)))
ox-->           (set! ydist (exact->inexact (- 150 j)))
ox-->           (set! dist (sqrt (+ (sqr xdist) (sqr ydist))))
ox-->           (cond
ox-->            ((< dist 61.0)
ox-->             (set! Rred (pip-pixel-kernal redch Rred blurvec3 i j))
ox-->             (set! Rgrn (pip-pixel-kernal grnch Rgrn blurvec3 i j))
ox-->             (set! Rblu (pip-pixel-kernal bluch Rblu blurvec3 i j))
ox-->             )
ox-->            )
ox-->           )
ox-->      )

ox--> (pip-show (list Rred Rgrn Rblu) 'label "10 Most Wanted")

Witless Protection Program

source image	regionally blurred image

PiP: The Proceedural Image Processing platform is Copyright © 1996,1997, Peter G. Carswell

Last updated: 02/08/97 / Peter Carswell ( pete@cgrg.ohio-state.edu )

Any comments or suggestions appreciated.