Feret angle funkiness

[eloprest]

Hello,

I apologize if this is a stupid question, but I am at the end of my rope with it.

I am in the process of standardizing photographs (of eggs) and I need all of the eggs to lie along the same axis, therefore, I have a macro written to take the feret angle and rotate the eggs by that much, which ought to orient them against the x-axis. To my eyes, this seems to work, they look like they are in the right position. However, I went back and calculated the feret angle of the selections and they are all off – albeit by just a little bit: .5 - 3 degrees, which is not huge, but certainly would throw off calculations.

I cannot think of a reason for this (as the pixels retain their orientation in relation to each other, right?) and am puzzled and hesitant to run analyses on the images. I searched the forums and could not find anyone else with the same problem. Should I be going about this in an entirely different way?

Thanks in advance,
Eric

[Gabriel Landini]

On Monday 05 Dec 2011 23:00:01 you wrote:

> to take the feret angle and rotate the eggs by that much, which ought to
> orient them against the x-axis. To my eyes, this seems to work, they look
> like they are in the right position. However, I went back and calculated
> the feret angle of the selections and they are all off – albeit by just a
> little bit: .5 - 3 degrees, which is not huge, but certainly would throw
> off calculations.
>
> I cannot think of a reason for this (as the pixels retain their orientation
> in relation to each other, right?) and am puzzled and hesitant to run
> analyses on the images. I searched the forums and could not find anyone
> else with the same problem. Should I be going about this in an entirely
> different way?

That is expected. The discrete digitised boundaries do not have the longest
axis where one expects they should be in their continuous version.
In *digitised* ellipses the computed Feret diameter is on the diagonal of a
inscribed rectangle (depending on the aspect ratio) rather than the medial
axis of that inscribed rectangle. Draw a small ellipse in IJ and see for
yourself. This is the prize one pays for digitising shapes.

Cheers
G.

[Norbert Vischer]

Hello Eric,

> to take the feret angle and rotate the eggs by that much, which ought to
> orient them against the x-axis. To my eyes, this seems to work, they look
> like they are in the right position. However, I went back and calculated
> the feret angle of the selections and they are all off – albeit by just a
> little bit: .5 - 3 degrees, which is not huge, but certainly would throw
> off calculations.
>
> I cannot think of a reason for this (as the pixels retain their orientation
> in relation to each other, right?) and am puzzled and hesitant to run
> analyses on the images. I searched the forums and could not find anyone
> else with the same problem. Should I be going about this in an entirely
> different way?

If the eggs are symmetrical about the long axis (what I periodically observe in the super market), you could use the angle that is calculated from the best-fitting ellipse. This value is based on all pixels of the object and not only to the perimeter, it suffers much less from pixelisation mentioned by Gabriel.

Norbert Vischer

[eloprest]

Gracias, amigos.

It makes good sense now that I see how it is calculated.

-Eric

On Mon, Dec 5, 2011 at 9:12 PM, Norbert Vischer <[hidden email]>wrote:

> Hello Eric,
>
> > to take the feret angle and rotate the eggs by that much, which ought to
> > orient them against the x-axis. To my eyes, this seems to work, they look
> > like they are in the right position. However, I went back and calculated
> > the feret angle of the selections and they are all off – albeit by just a
> > little bit: .5 - 3 degrees, which is not huge, but certainly would throw
> > off calculations.
> >
> > I cannot think of a reason for this (as the pixels retain their
> orientation
> > in relation to each other, right?) and am puzzled and hesitant to run
> > analyses on the images. I searched the forums and could not find anyone
> > else with the same problem. Should I be going about this in an entirely
> > different way?
>
> If the eggs are symmetrical about the long axis (what I periodically
> observe in the super market), you could use the angle that is calculated
> from the best-fitting ellipse. This value is based on all pixels of the
> object and not only to the perimeter, it suffers much less from
> pixelisation mentioned by Gabriel.
>
> Norbert Vischer
>