opinion about image analysis development from zero

[Constantin Frangoulis]

Dear Java users,

We are a team of biologist that would like to start on image analysis of plankton
(phytoplankton, microzooplankton and mesozooplankton). We know few things
about image analysis and nothing about Java. We would like to have your opinion
about a difficult choice we want to make

a)Buy a very expensive software (Image Pro Plus) with which we will have
technical support but we are afraid that could be limited for the applications we will have to
make
b) Buy expensive hardware (cameras, computer, screen etc) and use a free software that uses Java
 and is adapted to our needs but still under development (e.g. Phyto/Zooimage). In this case we are
 afraid that this will need a lot of time for us to learn and adapt.

Could you give us your opinion about this from your experience
Thank you

Best

Kosta

[Mikhail Umorin]

Kosta --

I think your priority should be camera hardware, because no software can
compensate for a noisy, low resolution image. I have tried to do some
computerized phytoplankton identification from images, and was stalled by
the poor
quality of images (taken by basically a colour TV camera). I did not get
beyond Chlorella :). I used Image Pro +
(several years ago) and ImageJ for identification. In my experience ImagePro
has a richer set of features (as relates to segmentation and counting) and a
more convenient integration of those features. ImageJ can do many of the same
things but menus and plugins are very scattered and you have to be creative
about how you put them together. You can achieve many things with macros but
to do more intricate things with images you will need to write Java plugins.
The missing features can certainly be added to ImageJ but they may require
some hardcore programming. But that can be learnt. When I needed stack
z-projection and earlier versions of ImagePro I worked with did not have
that feature, I just wrote a plugin (Stack Focuser) for ImageJ.  (Plankton
identification, though, will certainly be a much more challenging task.) More
recent ImagePro has stack projection but ImageJ has at least three (I think)
different plugins for the task. So, ImageJ is much more flexible but it takes
more effort to utilize it. Access to technical support for ImagePro may
not be of value to you because they will only help you to work with existing
features of the software. They will not write a new feature just for you. And
if they have many requests for it they will introduce it into a new version
and you will still have to buy it.

As for the challenges of image recognition itself you will need excellent
optics and a high-resolution low-noise camera. If specimens do not move then
you can get b/w camera (preferably cooled) and use different filters to get
RGB or multi-channel fluorescence images for relatively same cost as a
lower-end color camera. You will gain in higher resolution, lower noise,
higher sensitivity and greater versatility. If you plan to use only one
setting for image acquisition (say, only bright field at 40x, or only UV
fluorescence at 100x) then you may be able to lower your hardware costs with
some lower-end camera that nevertheless works well in you particular setup.
Specimen identification often involves detection of such fine morphological
structures as flagella, bristles, and stained cell inclusions; you will need
high resolution for that. Specimens often do not appear in focus over entire
specimen area so you will need to do either 3-d reconstruction or flattening
from a stack, or use smaller magnification which will give you greater depth
of field (and make specimens more in focus). In this case, a higher-res
camera will help in discerning morphological features on a specimen that on a
lower-res camera image would occupy only 10 or 20 pixels. Illumination and
optical setup are extremely important and must be stable and reproducible.
Evenness of background lack of color and geometrical aberrations, and flat
field will go a long way in saving you numerous image recognition pains
later.

That's what I think. Hope it helps,

Mikhail.

On Wed December 20 2006 05:26, Constantin Frangoulis wrote:

> Dear Java users,
>
> We are a team of biologist that would like to start on image analysis of
> plankton (phytoplankton, microzooplankton and mesozooplankton). We know few
> things about image analysis and nothing about Java. We would like to have
> your opinion about a difficult choice we want to make
>
> a)Buy a very expensive software (Image Pro Plus) with which we will have
> technical support but we are afraid that could be limited for the
> applications we will have to make
> b) Buy expensive hardware (cameras, computer, screen etc) and use a free
> software that uses Java and is adapted to our needs but still under
> development (e.g. Phyto/Zooimage). In this case we are afraid that this
> will need a lot of time for us to learn and adapt.
>
> Could you give us your opinion about this from your experience
> Thank you
>
> Best
>
> Kosta

[Chu, Calvin-2]

I certainly agree with Mikhail that it is definitely better to optimize your image quality with better image acquisition.  there is absolutely no way you can compensate for a bad image.  It will make your life a lot easier in the post-processing end if you have images of very good quality.

However, I have used both imagePro and imageJ, and I personally don't feel that imagePro is any better than imageJ.  Everything you can do in imagePro you can do in imageJ.  In addition, there are so many people that use imageJ it is probably very likely that someone else has done or is trying to do the exact same thing as you are.  Therefore you should utilize this forum as much as you can.  Furthermore, it is much easier to program something in imageJ than it is in imagePro.  The language used in imagePro is quite cryptic and my colleagues don't find it to be very reliable.  

The only reason my colleagues use imagePro is for the automated focusing, acquisition, and control of the microscope.  You can create some very fast automatic focus algorithms for high-throughput or large scale image acquisition.  

Lastly, you should keep in mind that the technical support for imagePro is horrible, and I have heard some terrible stories about configuring imagePro.

Calvin

 

Hope this helps

 

Kosta --

I think your priority should be camera hardware, because no software can
compensate for a noisy, low resolution image. I have tried to do some
computerized phytoplankton identification from images, and was stalled by
the poor
quality of images (taken by basically a colour TV camera). I did not get
beyond Chlorella :). I used Image Pro +
(several years ago) and ImageJ for identification. In my experience ImagePro
has a richer set of features (as relates to segmentation and counting) and a
more convenient integration of those features. ImageJ can do many of the same
things but menus and plugins are very scattered and you have to be creative
about how you put them together. You can achieve many things with macros but
to do more intricate things with images you will need to write Java plugins.
The missing features can certainly be added to ImageJ but they may require
some hardcore programming. But that can be learnt. When I needed stack
z-projection and earlier versions of ImagePro I worked with did not have
that feature, I just wrote a plugin (Stack Focuser) for ImageJ.  (Plankton
identification, though, will certainly be a much more challenging task.) More
recent ImagePro has stack projection but ImageJ has at least three (I think)
different plugins for the task. So, ImageJ is much more flexible but it takes
more effort to utilize it. Access to technical support for ImagePro may
not be of value to you because they will only help you to work with existing
features of the software. They will not write a new feature just for you. And
if they have many requests for it they will introduce it into a new version
and you will still have to buy it.

As for the challenges of image recognition itself you will need excellent
optics and a high-resolution low-noise camera. If specimens do not move then
you can get b/w camera (preferably cooled) and use different filters to get
RGB or multi-channel fluorescence images for relatively same cost as a
lower-end color camera. You will gain in higher resolution, lower noise,
higher sensitivity and greater versatility. If you plan to use only one
setting for image acquisition (say, only bright field at 40x, or only UV
fluorescence at 100x) then you may be able to lower your hardware costs with
some lower-end camera that nevertheless works well in you particular setup.
Specimen identification often involves detection of such fine morphological
structures as flagella, bristles, and stained cell inclusions; you will need
high resolution for that. Specimens often do not appear in focus over entire
specimen area so you will need to do either 3-d reconstruction or flattening
from a stack, or use smaller magnification which will give you greater depth
of field (and make specimens more in focus). In this case, a higher-res
camera will help in discerning morphological features on a specimen that on a
lower-res camera image would occupy only 10 or 20 pixels. Illumination and
optical setup are extremely important and must be stable and reproducible..
Evenness of background lack of color and geometrical aberrations, and flat
field will go a long way in saving you numerous image recognition pains
later.

That's what I think. Hope it helps,

Mikhail.

On Wed December 20 2006 05:26, Constantin Frangoulis wrote:

> Dear Java users,
>
> We are a team of biologist that would like to start on image analysis of
> plankton (phytoplankton, microzooplankton and mesozooplankton). We know few
> things about image analysis and nothing about Java. We would like to have
> your opinion about a difficult choice we want to make
>
> a)Buy a very expensive software (Image Pro Plus) with which we will have
> technical support but we are afraid that could be limited for the
> applications we will have to make
> b) Buy expensive hardware (cameras, computer, screen etc) and use a free
> software that uses Java and is adapted to our needs but still under
> development (e.g. Phyto/Zooimage). In this case we are afraid that this
> will need a lot of time for us to learn and adapt.
>
> Could you give us your opinion about this from your experience
> Thank you
>
> Best
>
> Kosta