[FFmpeg-devel] [PATCH v2] avfilter/vf_edgedetect: properly implement double_threshold()

[Andriy Gelman]

On Mon, 22. Jun 17:29, Valery Kot wrote:
> === Version 1 ===
> vf_edgedetect video filter implements Canny algorithm
> (https://en.wikipedia.org/wiki/Canny_edge_detector)
> 
> Important part of this algo is the double threshold step: pixels above
> "high" threshold being kept, pixels below "low" threshold dropped,
> pixels in between kept if they are attached to "high" pixels.
> 
> This is implemented in the double_threshold() function. However,
> condition to start checking attached pixels, as it is now and as it
> was in FFmpeg since 2012, only allows checking on the boundary, not
> inside the video. It is a very lucky coincidence that those boundary
> pixels are always 0, otherwise following lines would be reading
> outside of the buffer.
> 
> As it is now, double_threshold() only implements "high" thresholding.
> As a result, edges are either noisy or fragmented, depending on "high"
> threshold selection; "low" threshold is simply ignored.
> 
> Attached one char patch fixes this.
> 
> === Version 2 ===
> - include avfilter/ in commit message
> - update FATE tests

> From 69bbe24bfe23efa3874448f28451b1abaa209d5d Mon Sep 17 00:00:00 2001
> From: vkot <valery.kot at kinetiq.tv>
> Date: Fri, 19 Jun 2020 16:57:13 +0200
> Subject: [PATCH] avfilter/vf_edgedetect: properly implement double_threshold()
> 
> ---
>  libavfilter/vf_edgedetect.c               | 2 +-
>  tests/ref/fate/filter-edgedetect          | 2 +-
>  tests/ref/fate/filter-edgedetect-colormix | 2 +-
>  3 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/libavfilter/vf_edgedetect.c b/libavfilter/vf_edgedetect.c
> index a5614ea63b..df8afbd532 100644
> --- a/libavfilter/vf_edgedetect.c
> +++ b/libavfilter/vf_edgedetect.c
> @@ -294,7 +294,7 @@ static void double_threshold(int low, int high, int w, int h,
>                  continue;
>              }
>  
> -            if ((!i || i == w - 1 || !j || j == h - 1) &&
> +            if (!(!i || i == w - 1 || !j || j == h - 1) &&

lgtm. I saw a small improvement when testing.

Would be nice to allow weak edges that become strong to trigger neighboring weak
edges in the future.

Thanks,
-- 
Andriy