[GRASS-user] Running canny edge detection on thinned binary image

Venka · July 24, 2023, 3:28am

Hi All,

I have a question about Canny Edge Detection using i.edge

1) I produce a thinned binary image as below

r.thin --overwrite input=line_element output=thinned_line_element

a) the "line_element" represents valleys that are, at places, few pixel
wide

b) the output "thinned_line_element" represents valley lines that are
a single pixel wide

2) I run the Canny edge detector on "thinned_line_element" as below
i.edge --overwrite input=thin_line_element output=edge_map angles_map=angle_map

The resultant "edge_map" produces a monotone (no edges) edge map and the "angle_map" outputs the angles
correctly

3) Running the Canny edge detector on the un-thinned "line_element"

i.edge --overwrite input=line_element@PERMANENT output=edge_map1 angles_map=angle_map1

produces both "edge_map1" and "angles_map1" correctly.

Any idea why the Canny edge detector does not produce edge map on thinned image?

Kind regards,

Venka

Micha_Silver1 · July 24, 2023, 9:23am

AFAIK, the Canny algorithm requires large, wide areas of black pixels, adjacent to large wide areas of white pixels in order to find the edge. It compares the change in value over a wide “strip” to determine the edge.

It’s a bit counter-intuitive, but the algorithm will not work on single pixel width areas.

···

On 24/07/2023 6:28, Venka wrote:

Hi All,

I have a question about Canny Edge Detection using i.edge

I produce a thinned binary image as below

r.thin --overwrite input=line_element output=thinned_line_element

a) the “line_element” represents valleys that are, at places, few pixel
wide

b) the output “thinned_line_element” represents valley lines that are
a single pixel wide

I run the Canny edge detector on “thinned_line_element” as below
i.edge --overwrite input=thin_line_element output=edge_map angles_map=angle_map

The resultant “edge_map” produces a monotone (no edges) edge map and the “angle_map” outputs the angles
correctly

Running the Canny edge detector on the un-thinned “line_element”

i.edge --overwrite input=line_element@PERMANENT output=edge_map1 angles_map=angle_map1

produces both “edge_map1” and “angles_map1” correctly.

Any idea why the Canny edge detector does not produce edge map on thinned image?

Kind regards,

Venka

grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user

-- 
Micha Silver
Ben Gurion Univ.
Sde Boker, Remote Sensing Lab
cell: +972-523-665918

Venka · July 24, 2023, 10:18am

Hi Micha,

Thanks for your response.

Is there any other way to generate edge_map and angle_map
from single pixel width areas?

Venka

On 7/24/2023 6:23 PM, Micha Silver wrote:

AFAIK, the Canny algorithm requires large, wide areas of black pixels, adjacent
to large wide areas of white pixels in order to find the edge. It compares the
change in value over a wide "strip" to determine the edge.

It's a bit counter-intuitive, but the algorithm will not work on single pixel
width areas.

On 24/07/2023 6:28, Venka wrote:

Hi All,

I have a question about Canny Edge Detection using i.edge

1) I produce a thinned binary image as below

r.thin --overwrite input=line_element output=thinned_line_element

a) the "line_element" represents valleys that are, at places, few pixel
wide

b) the output "thinned_line_element" represents valley lines that are
a single pixel wide

2) I run the Canny edge detector on "thinned_line_element" as below
i.edge --overwrite input=thin_line_element output=edge_map angles_map=angle_map

The resultant "edge_map" produces a monotone (no edges) edge map and the "angle_map" outputs the angles
correctly

3) Running the Canny edge detector on the un-thinned "line_element"

i.edge --overwrite input=line_element@PERMANENT output=edge_map1 angles_map=angle_map1

produces both "edge_map1" and "angles_map1" correctly.

Any idea why the Canny edge detector does not produce edge map on thinned image?

Kind regards,

Venka
_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user

--
Micha Silver
Ben Gurion Univ.
Sde Boker, Remote Sensing Lab
cell: +972-523-665918

wenzeslaus · July 25, 2023, 1:11am

Hi Venka,

i.edge, or Canny Edge Detector, is definitely built for something else. Lines are output, not input. However, we created i.edge together with r.houghtransform which works on the result from i.edge or any other thin lines. The line segments it extracts might be what you are looking for. Maybe you are just looking for r.to.vect followed by v.to.db with azimuth, maybe with v.split in between.

Best,
Vaclav

On Mon, 24 Jul 2023 at 06:18, Venka <venka.osgeo@gmail.com> wrote:

Hi Micha,

Thanks for your response.

Is there any other way to generate edge_map and angle_map
from single pixel width areas?

Venka

On 7/24/2023 6:23 PM, Micha Silver wrote:

AFAIK, the Canny algorithm requires large, wide areas of black pixels, adjacent
to large wide areas of white pixels in order to find the edge. It compares the
change in value over a wide “strip” to determine the edge.

It’s a bit counter-intuitive, but the algorithm will not work on single pixel
width areas.

On 24/07/2023 6:28, Venka wrote:

Hi All,

I have a question about Canny Edge Detection using i.edge

I produce a thinned binary image as below

r.thin --overwrite input=line_element output=thinned_line_element

a) the “line_element” represents valleys that are, at places, few pixel
wide

b) the output “thinned_line_element” represents valley lines that are
a single pixel wide

I run the Canny edge detector on “thinned_line_element” as below
i.edge --overwrite input=thin_line_element output=edge_map angles_map=angle_map

The resultant “edge_map” produces a monotone (no edges) edge map and the
“angle_map” outputs the angles
correctly

Running the Canny edge detector on the un-thinned “line_element”

i.edge --overwrite input=line_element@PERMANENT output=edge_map1
angles_map=angle_map1

produces both “edge_map1” and “angles_map1” correctly.

Any idea why the Canny edge detector does not produce edge map on thinned image?

Kind regards,

Venka

grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user

–
Micha Silver
Ben Gurion Univ.
Sde Boker, Remote Sensing Lab
cell: +972-523-665918

grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user

Venka · July 25, 2023, 1:55am

Hi Vaclav,

Thanks for your response.

My purpose is to generate edge and angle map and use them as
input for r.houghtransform.

The workflow works well for detecting linear valley features
form DEM. I notice that r.houghtransform works better when we
use the angle map and set the appropriate angle_width parameter
as shown below.

r.houghtransform input=edge_map@PERMANENT output=line_map angles=angle_map@PERMANENT angle_width=6 lines_number=200 gap_size=3 max_gap_count=6 max_gap=6 line_width=4 --overwrite

I was trying to find a way to generate angle map for a thinned
"line element" image and use the "line element" image and angle
map to extract lines using r.houghtransform.

Thanks for pointing to v.to.db with azimuth option, I will try to figure
out if it can be use to generate an angle map.

Best,

Venka

On 7/25/2023 10:11 AM, Vaclav Petras wrote:

Hi Venka,

i.edge, or Canny Edge Detector, is definitely built for something else.
Lines are output, not input. However, we created i.edge together with
r.houghtransform which works on the result from i.edge or any other thin
lines. The line segments it extracts might be what you are looking for.
Maybe you are just looking for r.to.vect followed by v.to.db with azimuth,
maybe with v.split in between.

Best,
Vaclav

On Mon, 24 Jul 2023 at 06:18, Venka <venka.osgeo@gmail.com> wrote:

Hi Micha,

Thanks for your response.

Is there any other way to generate edge_map and angle_map
from single pixel width areas?

Venka

On 7/24/2023 6:23 PM, Micha Silver wrote:

AFAIK, the Canny algorithm requires large, wide areas of black pixels,

adjacent

to large wide areas of white pixels in order to find the edge. It

compares the

change in value over a wide "strip" to determine the edge.

It's a bit counter-intuitive, but the algorithm will not work on single

pixel

width areas.

On 24/07/2023 6:28, Venka wrote:

Hi All,

I have a question about Canny Edge Detection using i.edge

1) I produce a thinned binary image as below

r.thin --overwrite input=line_element output=thinned_line_element

a) the "line_element" represents valleys that are, at places, few pixel
wide

b) the output "thinned_line_element" represents valley lines that are
a single pixel wide

2) I run the Canny edge detector on "thinned_line_element" as below
i.edge --overwrite input=thin_line_element output=edge_map

angles_map=angle_map

The resultant "edge_map" produces a monotone (no edges) edge map and

the

"angle_map" outputs the angles
correctly

3) Running the Canny edge detector on the un-thinned "line_element"

i.edge --overwrite input=line_element@PERMANENT output=edge_map1
angles_map=angle_map1

produces both "edge_map1" and "angles_map1" correctly.

Any idea why the Canny edge detector does not produce edge map on

thinned image?

Kind regards,

Venka
_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user

--
Micha Silver
Ben Gurion Univ.
Sde Boker, Remote Sensing Lab
cell: +972-523-665918

_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-user