I have a tif image of size around ~10 Gb. I need to perform object classification or pixel classification in this image. The dimension of image data has zyx form. My voxel size in x=0.6, y=0.6 and z=1.2.
Z is the depth of the object.
If I do classification of pixels in each Z plane separately and then merge to get the final shape and volume of object. Would I loose any information and my final shape or volume of object will be wrong?
If the classification of a pixel depends on the neighboring pixels (a gradient for example), then it is better to use the information of the neighboring out of plane pixels as well.
Generally speaking, if you have a 3D problem and have 3D data available, use all available data!
You also might avoid topological problems. For example, if you have segmented cell boundaries in each slice, those contours can be connected to contours in other slices, in many many different ways. Marching cubes in 3D would avoid such problems.
Only if the classification is based on the pixel-value itself (thresholding), only then you can treat each slice separately.
Thank you very much for your answer. I got the point. The problem is that using full 3d data Ilastik or ImageJ softwere becoming very very slow and I can't use it for such large dataset. What would you recommend for such large dataset? Shall I convert Tiff file to Jpeg image sequence and then I proceed?
Converting to jpg will not help: in order to do image processing, the jpg needs to be decompressed so you gain nothing and get compression artifacts for free!
- If you don't need the full resolution, you could downsize the xy planes by a factor two. This has the advantage that the resolution along the three axes is then equal, and the volume size is reduced by a factor four.
- Maybe you can reduce the dynamic range of the pixels. Are they color or grey or 16 bits or what?
- You might use sub-volumes.
- Invest in hardware...
Thanks. My image is color and 16 bit. I will reduce the resolution and then try to do segmentation. Sub-volume idea is also good.
Thanks a lot.
You may try some transformation (IWT, DWT etc. or any suitable for your data) to avoid possible losses (in down sizing) and make the analysis somewhat easy.
If your data have a very high slice spacing in 3rd dimension and you are getting 2d slices results, you need to integrate neighbouring slices, with a fixed spacing of 1 mm using a linear interpolation to deal with this problem.
Mubashir Ahmad , why does one need to interpolate in-between slices? The interpolated slices do not add information; they are just fantasy, and cause blurring in the interpolated dimension. The only reason I can think of is that surface rendering of 3D structures look less bumpy, but linear interpolation does not solve that. Cubic spline or Catmull-Rom interpolation are better choices.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification