Skip banner and top navigation
Go to Site Home (Online Learning Tool for Research Integrity and Image Processing)  
Guideline #7 - Filters Degrade Data

Filters Degrade Data

Use of software filters to improve image quality is usually not recommended for biological images.

Cartoon for Guideline7
  • Commercial digital image manipulation software was primarily written for graphics specialists in the design and print industries, not for scientists.  Software filters found in commercial software cannot be trusted to appropriately manipulate scientific image data.
  • Software-based digital image filters are mathematical functions called convolution kernels.  To see how a convolution kernel works, see this web page at the Florida State University Molecular Expressions website (Suggestion - press the AUTO button).
  • Software filters typically change the numerical intensity value of every pixel in the image.
    • Convolution kernels perform their mathematical functions using the numerical intensity values in small regions of the image (typically 3x3, 5x5, 7x7 pixel areas).
    • This means that the amount of change to the intensity of an individual pixel that is caused by the convolution kernel can be different in different parts of the image.  While the mathematical function is uniform, its effect is different in different parts of an image.
    • Under certain circumstances, convolution kernels can create an artifact or multiple artifacts in an image.  An artifact is defined as “a structure or substance not normally present but produced by an external agent or action...1
    • If users do not carefully compare a filtered image with the original image, it is possible that artifacts could be incorrectly interpreted as meaningful data.
  • If software filters must be used on scientific image data, the filters should be noted in an article’s figure legends or methods section.  The notation should include the software version, filter name(s) and any special settings that were used.
  • On a related note: software filters (guideline #7) and to some extent the techniques in guidelines #6 (selective enhancement) and #8 (cloning and copying) have sometimes been used to “clean up” the background in images.  Depending on the sample, backgrounds may be present in a digital image due to; non-specific staining, a less-than pristine preparatory technique (dirt) or electronic noise from the detector.
  • Scientists should keep in mind the possibility that someone will look at their data in a way they had not considered.  Perhaps the collagen matrix, support media, interface between two structures, or other “unimportant” features in the image contain information that will spark an idea for a reader’s research.  If the author changes the “unimportant” parts of an image to enhance the aspects in the image that they think are important, the author has lied to the reader and possibly removed the opportunity for a serendipitous finding.
  • “Data beautification”2 is a form of misrepresentation even when it doesn’t completely cross the line into outright falsification

Skip footer links and go to content