ImageJ

Automatized object-identification / segmentation of fish from photos

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

5 messages Options

Johannes Radinger

Automatized object-identification / segmentation of fish from photos

Hi ImageJ users,

I am very new to ImageJ and not at all familiar with the capabilities of
this tool-set. What I'd actually like to do is to extract a fish from a
photo, i.e. separate the fish from its background. And as I want to do this
for many images I am looking for an automatized way. The images look like:

http://fishbase.org/photos/PicturesSummary.php?ID=4730&what=species
http://fishbase.org/photos/PicturesSummary.php?StartRow=0&ID=4662&what=species&TotRec=5

Is it generally possible to detect the fish as an object and to select it.
I thought about some kind of image segmentation, however the the background
is not uniform which might complicate things. Two things are fixed: 1)
There is always only one fish, so the final result should contain one fish
and one background 2) The fish is always more or less centered within the
image.

Any ideas?

/Johannes

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Herbie

Re: Automatized object-identification / segmentation of fish from photos

Good day Johannes,

you may be lucky with some of your images but as you write:

"[...] however the the background is not uniform which might complicate
things"

Segmentation will not be satisfying in most cases, with images such as
the second example image.

Please note that it is very hard and sometimes even impossible to
remedie problems by image processing and analysis that are due to
insufficient image acquisition.

Why not isolate every fish in a small aqurium with suitable background?

Regards

Herbie

:::::::::::::::::::::::::::::::::::::::::::::::
Am 26.06.17 um 12:31 schrieb Johannes Radinger:

> Hi ImageJ users,
>
> I am very new to ImageJ and not at all familiar with the capabilities of
> this tool-set. What I'd actually like to do is to extract a fish from a
> photo, i.e. separate the fish from its background. And as I want to do this
> for many images I am looking for an automatized way. The images look like:
>
> http://fishbase.org/photos/PicturesSummary.php?ID=4730&what=species
> http://fishbase.org/photos/PicturesSummary.php?StartRow=0&ID=4662&what=species&TotRec=5
>
> Is it generally possible to detect the fish as an object and to select it.
> I thought about some kind of image segmentation, however the the background
> is not uniform which might complicate things. Two things are fixed: 1)
> There is always only one fish, so the final result should contain one fish
> and one background 2) The fish is always more or less centered within the
> image.
>
> Any ideas?
>
> /Johannes
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Jeremy Adler-2

Re: Automatized object-identification / segmentation of fish from photos

Johannes & Herbie,
Segmentation is clearly very difficult.

But presumably the fish move - it should then be possible to acquire a second image without the fish.
The difference between the two images is - the fish.

-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Herbie
Sent: den 26 juni 2017 14:04
To: [hidden email]
Subject: Re: Automatized object-identification / segmentation of fish from photos

Good day Johannes,

you may be lucky with some of your images but as you write:

"[...] however the the background is not uniform which might complicate things"

Segmentation will not be satisfying in most cases, with images such as the second example image.

Please note that it is very hard and sometimes even impossible to remedie problems by image processing and analysis that are due to insufficient image acquisition.

Why not isolate every fish in a small aqurium with suitable background?

Regards

Herbie

:::::::::::::::::::::::::::::::::::::::::::::::
Am 26.06.17 um 12:31 schrieb Johannes Radinger:

> Hi ImageJ users,
>
> I am very new to ImageJ and not at all familiar with the capabilities
> of this tool-set. What I'd actually like to do is to extract a fish
> from a photo, i.e. separate the fish from its background. And as I
> want to do this for many images I am looking for an automatized way. The images look like:
>
> http://fishbase.org/photos/PicturesSummary.php?ID=4730&what=species
> http://fishbase.org/photos/PicturesSummary.php?StartRow=0&ID=4662&what
> =species&TotRec=5
>
> Is it generally possible to detect the fish as an object and to select it.
> I thought about some kind of image segmentation, however the the
> background is not uniform which might complicate things. Two things
> are fixed: 1) There is always only one fish, so the final result
> should contain one fish and one background 2) The fish is always more
> or less centered within the image.
>
> Any ideas?
>
> /Johannes
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Johannes Radinger

Re: Automatized object-identification / segmentation of fish from photos

FYI, actually, I am not taking the pictures myself, I'll rather use already
taken pictures e.g. as from fishbase.org. So I don't have the possibility
to select my aquarium background or to take to independent pictures (w and
w/o the fish)...
Anyway I'll give ilastik a try and read deeper into DNN, but I guess an
automatized way might be very difficult not to say impossible.

/johannes

On Mon, Jun 26, 2017 at 2:11 PM, Jeremy Adler <[hidden email]
> wrote:

> Johannes & Herbie,
> Segmentation is clearly very difficult.
>
> But presumably the fish move - it should then be possible to acquire a
> second image without the fish.
> The difference between the two images is - the fish.
>
>
> -----Original Message-----
> From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
> Herbie
> Sent: den 26 juni 2017 14:04
> To: [hidden email]
> Subject: Re: Automatized object-identification / segmentation of fish from
> photos
>
> Good day Johannes,
>
> you may be lucky with some of your images but as you write:
>
> "[...] however the the background is not uniform which might complicate
> things"
>
> Segmentation will not be satisfying in most cases, with images such as the
> second example image.
>
> Please note that it is very hard and sometimes even impossible to remedie
> problems by image processing and analysis that are due to insufficient
> image acquisition.
>
> Why not isolate every fish in a small aqurium with suitable background?
>
> Regards
>
> Herbie
>
> :::::::::::::::::::::::::::::::::::::::::::::::
> Am 26.06.17 um 12:31 schrieb Johannes Radinger:
> > Hi ImageJ users,
> >
> > I am very new to ImageJ and not at all familiar with the capabilities
> > of this tool-set. What I'd actually like to do is to extract a fish
> > from a photo, i.e. separate the fish from its background. And as I
> > want to do this for many images I am looking for an automatized way. The
> images look like:
> >
> > http://fishbase.org/photos/PicturesSummary.php?ID=4730&what=species
> > http://fishbase.org/photos/PicturesSummary.php?StartRow=0&ID=4662&what
> > =species&TotRec=5
> >
> > Is it generally possible to detect the fish as an object and to select
> it.
> > I thought about some kind of image segmentation, however the the
> > background is not uniform which might complicate things. Two things
> > are fixed: 1) There is always only one fish, so the final result
> > should contain one fish and one background 2) The fish is always more
> > or less centered within the image.
> >
> > Any ideas?
> >
> > /Johannes
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> >
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Robert Smith

Re: Automatized object-identification / segmentation of fish from photos

Radinger ,

Actually it may be much easier to get a video of whatever it is you may want (e.g. AVI) as an image sequence and then us it as a stacked image. Then use the individual images to get the one you want or one (or several) to perform the math required.

I would be difficult to remove one image from another with much work.

Try it. If you need more advice on this technique let me know.

Bob Smith

________________________________
From: Johannes Radinger <[hidden email]>
Sent: Monday, June 26, 2017 8:23 AM
To: [hidden email]
Subject: Re: Automatized object-identification / segmentation of fish from photos

FYI, actually, I am not taking the pictures myself, I'll rather use already
taken pictures e.g. as from fishbase.org. So I don't have the possibility
to select my aquarium background or to take to independent pictures (w and
w/o the fish)...
Anyway I'll give ilastik a try and read deeper into DNN, but I guess an
automatized way might be very difficult not to say impossible.

/johannes

On Mon, Jun 26, 2017 at 2:11 PM, Jeremy Adler <[hidden email]
> wrote:

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html