I am using Image J to measure the length/width (freehand tool), but I am not sure whether I should be measuring areas instead? This has been tricky to measure as it's more stepwise process (making a binary image, set threshold, and then select each abdomen area). I have several hundreds of individual bees (5 per photo) to measure. So any tip to analyse this accurately and faster would be more than welcome. Thank you.
If you include a single example of the images you captured that would be helpful to me or others to answer your Q.
I would like to strongly suggest you take a look at our newly released software MIPAR, available as an extended free trial at http://MIPAR.us. It's recipe creation feature in the Image Processor, ability to execute the recipe on multiple images in the Batch Processor, and ability to rapidly review, edit, and measure all images in the Processed Image Editor would likely greatly accelerate your work. As Abbas mentioned, if you're willing to post an example image (or submit one at http://MIPAR.us/try-image), I'd be happy to try to make a recipe file and send it to you.
Cheers,
John
Thanks for posting - I'll get to work on a recipe! I can also post a screen recording of how to quickly flip through multiple images and outline each abdomen to make fast measurements if you'd like? I think this semi-auto approach might be the most efficient for you given the complexity of your segmentation problem, but I'll also take a stab at a recipe.
Hi Raquel,
Thanks for posting the image. I can see the problem and it is indeed a very challenging one. Even with the manual measurements I am afraid that you would not hope for so much precision for the following reasons:
A) Although all of the 5 bees are within the FOV, the sensor is not perpendicular to each one of them which causes geometric distortions. For example see the pinheads which I marked in the image, they are for sure of the same area but appear otherwise if you see my point.
B) Some bees appear to be imaged from their profile view while others are a bit leaned. This could also be the result of (A).
A 3D scanning would definitely be ideal but I guess it is not an option for you. I was really hopping to give you a better and more useful response but hopefully someone will post a good solution which will be interesting to know.
Hi Raquel,
I'm pretty excited about this! First, I have attached the roughly found abdomen from a fully automated MIPAR recipe. More importantly, I have attached a brief screen recording demonstrating the running of this recipe in MIPAR, followed by how to quickly edit the abdomen to improve accuracy, and finally color each according to size.
For a batch of images, I'd recommend running this recipe on all images, open the results in the Processed Image Editor, flip through each to make your edits, and then measure the abdomen sizes for all images at once.
I have also attached this recipe file. If you'd like to download MIPAR, please register at http://MIPAR.us/trial and I'll send a license key right away. I think this could dramatically accelerate your workflow and hope I can help!
Cheers,
John
As I understand from Raquel it is the precision and the full automation (100s images!!) of the process that she is after. Doing that manually using ImageJ could be possible following the steps highlighted in the attached file.
Hello,
you can use a simple way and instead of making one image you do it in stereo and use photogrammetry to measure 3D points on bees. The instrumental phase depends about the needed accuracy. Photogrammetry is easy to implement and you can choose the relevant density of points according to your needs. Two points for a length, thousand for a surface... For example I already measured fish size in situ, indeed you can measure accurately everything you can see .. :) If you need more info I can give you more details on the procedure ...
Thank you very much for your thoughts on this matters and suggestions. These measurements are not my main results. I am looking at taste preferences in bees so I measured food consumption given a choice of diet over time. And because I am using live animals I try to take as much information as I can. And so, more than food consumption and survival metrics I also measured the fresh weight of each bee (~50 per treatment, and I have several treatments..) at the end of the experiment. I started doing this as I realised that some bees (all the same age) would have very large abdomens compared to other in the same cohort, age, food availability etc..(this can also be a behavioural aspect). Thus, I decided to weight the bees across treatments.
Because I still had the bees why not checking for a correlation between weight and abdomen size. First I decided to shoot a bee sideways instead of backwards because of the distortion at the tip of the abdomen (sting). It seems to be that pinning a bee sideways I could trace an imaginary line between the tergites and sternites. I still have a curvature, but in the majority of the cases, the tip is in my FOV. I also started shooting 1 single with a macro lens and flash, but I have hundreds of them and I figure that displaying 5 of them per pic would still be on my main FOV and speed up the shooting process. Then I started measuring the length and width of each bee (around the 3rd tergite). Still some bees were not that easy to pin. So I am already accounting for some discrepancies, but that's why I collected 5 per box up to a total of 50 bees so I could check a trend/correlation between these 3 metrics (weight, abdomen size (length and width) and further with diet consumption. Furthermore, due to the differences between bees and pinning I came to the idea that perhaps measuring the abdomen area would be a better approach due to these distortions... But again and because I have other sorts of data to analyse it would be useful a faster way to measure these images.
Raquel, we understand that the measurements are not your ultimate goal after all. But their accuracy will indirectly affect your study's conclusions, don't you think so? A slight variability might not affect negatively your model since, as I would expect, your other statistical variables are more stable.
True, yes, and I do want things properly done (and inferred). I will have a go with some of your suggestions and I reckon with time the process will be better laborious but fast. Thank you again, Abbas!
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing