Segmentation Challenges and Definitions: Background ≠ Noise ≠ Unspecific Signal ≠ Uneven Illumination
Definitions of Unspecific Signals:

Biological image have a natural "background" consisting of different types of unspecific signals. Those unspecific signals listed below can have multiple negative influence during image analysis:

  • Imaging Background:
    • wrong instensity measurements, since the measured signal is too high
    • relative intensity comparison result in lower ratios / differences and therefore lead to less significant differences
  • Uneven Illumination / Lighting:
    • can lead to incorrect or missing objects detection
    • has a negative impact on object size and shape accuracy
    • leads to wrong intra-image (inside one image) intensity measurements
  • Unspecific Staining:
    • negative impact on accurate object detection
    • has a negative impact on object size and shape accuracy
    • can impact accuracy in intensity measurements
  • Auto Fluorescence:
    • increases inaccuracy in intensity measurements and relative intensity comparison
    • is very different in different images and image regions in one image and CANNOT be subtracted with simple methods!
  • Improper White Balance in color images:
    • negative impact on qualitative comparison of colors between images
    • can negatively influence area segmentation
  • Noise:
    • increases inaccuracy in intensity measurements
    • increases inaccuracy in object detection and shape

Reasons leading to unspecific signal and background should be addressed during sample preparation and image acquisition already. When corrected post-priori (after image acquisition) there are always assumptions involved which might be inaccurate or not determinable. Furthermore, most methods reduce those "undesired signals" artificially and might negatively impact the final intensity measurements. Therefore, correction methods have their limitations and need to be used with care. Intensity measurements should be done on original images on which necessary lighting correction and quantitative background subtraction is done, only!

One important consideration for unspecificity reduction mostly is that the more homogenious the unspecific signal is, the better the reduction will work.

Therefore, performing it after image filtering is mostly recommendable, since image filters normally homogenize the complete image content including background areas.