This plug-in performs various types of intensity projections along depth (for 3D data) or time (for timelapse data).
The principle of intensity projection is to project all slices (or time frames) of a sequence into a single 2D image. Each pixel of this final image is therefore a combination of all pixels with the same 2D coordinates in every projected image. The way theses pixel values are combined is determined by the selected algorithm.
5 projection algorithms are currently available, with a screenshot of what they produce on a sample (3D) data set (The brightness/contrast has been adjusted for purpose of demonstration, with the careful attempt of not saturating the final screenshot). Images used here are courtesy of Romain Thibeaux and Nancy Guillén (Institut Pasteur), and are not usable for reproduction without prior authorization.
First, here is one slice of the original 3D dataset. You won't see much, as it's pretty thick. That's just what we need to try out the various projection algorithms.
- Maximum Intensity Projection (a.k.a. MIP):
Each final projected pixel contains the maximum value of all pixels of same 2D coordinates in the original data. This algorithm is highly popular in 3D data visualization, as it gives a nice 2D glimpse of a potentially large data set. Note however that this algorithm has several issues: a) on an otherwise dark background, a single bright pixel that is due to imaging artifacts may appear in the final projection while it may not correspond to relevant information; b) on saturated data sets, the projection becomes less relevant, as it ends up mostly... saturated!
- Mean Intensity Projection:
Instead of the maximum above, each pixel holds the average value of all equivalent pixels in the data set. This algorithm is rarely used, as the information gets both dimmed and blurred (similar to a large mean filter along the projection axis). While this is good to remove the background noise, the structures of interest suffer as well.
- Median Intensity Projection:
Same as the mean above, except the average value is replaced by the median, with even stronger dimming and blurring effects (small structures of interest are considered as background noise, and therefore are entirely removed). Note that this projection does provide an efficient way to estimate the overall background.
- Standard Deviation Intensity Projection:
If you are at times dissatisfied with the maximum projection, try this one out. The standard deviation measure solves many of the previous issues: a) unlike the maximum (which select a single value), the standard deviation incorporates data from all slices/frames, and therefore handles saturated data pretty well (inner structures become visible again); b) by definition, the standard deviation does not average the data, it actually indicates the constrast variability across the data set, which is what we expect from a projected view after all.
- Saturated Sum:
Summing all pixel values usually has little interest (it quickly ends up saturating the resulting projection), this algorithm gives the same color dynamic as the mean projection on non-saturated projections. What's the point then? Well the idea is that saturation here won't necessary occur in every pixel, so there are regions where some details can be enhanced, notably in dim areas where the mean projection can't help.