Operators/Tools/Interactive/LabelImageEditor

From MiToBo
Revision as of 16:14, 4 February 2020 by Moeller (talk | contribs)
Jump to navigationJump to search



Label Image Editor

LabelImageEditor-MTBOperatorControlFrame.png

The Label Image Editor is available since release version 1.8.8 of MiToBo.
In MiToBo/MiToBo plugins release 1.8.13.1 it got a rich feature update.

Latest News

The editor got latest updates in MiToBo/MiToBo plugins 1.8.17.

Name of Plugin/Operator

de.unihalle.informatik.MiToBo.tools.interactive.LabelImageEditor
(available since MiToBo version 1.8.8)

Main features
  • images provided in an input folder are subsequently displayed to the user
  • regions, i.e. connected components, can be removed and joined by simple mouse-clicks
  • additional boundaries can be added by drawing free-hand lines
  • region labels can be adjusted to enhance visibility
  • region re-labeling is supported to guarantee unique labels
  • holes within regions can be filled
Usage

The editor is dedicated to the analysis of label images. In a label image each region or connected component, respectively, is assumed to be marked with a unique intensity or color value, i.e. all of its pixels should share this value. The editor provides various editing functions to remove or join regions within such images.

Note that the editor can also handle label images with one-pixel wide black boundaries between regions.

The editor takes as input a directory from where the images are loaded one after the other.
Once an image is displayed to the user edit operations can be performed. Afterwards the changes are saved to a new image file and the next label image is read from the given input directory.


To run the LabelImageEditor perform the following steps:

  • install MiToBo by following the instructions on the Installation page
  • run MiToBo and start the operator runner by selecting the menu item MiToBo Runner from Plugins -> MiToBo
  • in the selection menu navigate to de.unihalle.informatik.MiToBo.tool.interactive and select the operator LabelImageEditor

This will bring up the operator window of the LabelImageEditor (see figure on the right, top) where the basic configuration parameters can be added. Clicking the "Run" button will open the editor window with the first image loaded (figure on the right, bottom).


  • Input data:
    The operator reads the contents of a given input image directory and displays all images one after the other to the user.
    The user can edit the images, i.e. remove labeled regions, by clicking with the left mouse button somewhere into the region to remove.
    Clicking the "Next" button at the top of the editor window will save the current image to disk and proceed with the next image in the folder.

  • Output data:
    For each input image the operator will save a corresponding output image, either in the same folder or a specific output folder if provided.
    Output images will share the name of the corresponding input image extended with substring "-edited".
    Note, all images are relabeled automatically before saving to ensure consecutive labels in the resulting edited images.
    Consecutive labels are a pre-requisite for some other MiToBo tools which can be used in further processing steps, like the FeatureColorMapper.


  • Parameters:
Name Description
Input Directory directory from where the images to process will be loaded
File Filter Optional string to filter image files to be processed. If the string is empty all files are processed.
If a non-empty string is provided, files with names not containing the pattern will be skipped.
For example, if the string ".tif" is provided, only images ending with ".tif", i.e. in TIFF format, will be considered.
The string "cell" would select only images containing the word "cell" somewhere in their names.
Note that as string any valid Java regular expression is admissible. See for example this website for more details on regular expressions.
Output directory optional; edited result label images will be saved to this directory; if not provided the result images will be stored in the input folder
LabelImageEditor-EditorWindow.png


The following table lists all available operations available in the editor window:

Operation How to do it
remove a region click with left mouse button somewhere into the region to be removed;
the region will then be removed by setting all its pixels to a value of zero
join two regions press 'Shift' and left-click into the first region to remember its label, then release 'Shift' and left-click into the second region;
the label of the first region will be transferred to the second region; black pixels between both parts can be removed afterwards using the "Fix borders" button
draw a new boundary keep left mouse button pressed and drag the mouse; this will add a one-pixel wide free-hand line to the label image


The buttons in the editor window have the following functions:

Button Function Keyboard Shortcut
Next save current changes and jump to next image Alt + n
Skip cancel current image, don't save anything and jump to next image Alt + s
Contrast adjust labels to optimize visibility, i.e., shift labels into upper third of gray-scale range with maximal distances between them Alt + c
Relabel relabel the image, i.e., make sure that every region gets a unique label Alt + r
Fix Borders remove all background pixels with more than one foreground label in their 3x3 neighborhood Alt + b
Fill Holes fill background regions enclosed by a single region with the region label Alt + h
Undo undo the last action; IMPORTANT: only the most recent action can be undone! Alt + u
Quit save current changes and quit the editor window Alt + q



Hint: The Undo function has currently only a history of length one. If you accidentally deleted too many regions or want to undo more than one action, either run the editor once again on the complete directory or copy the corresponding image to a separate folder and run the editor on that folder. Results of a former run will be overwritten without further inquiry.

Updates



January 2019

  • The editor got a comprehensive update in MiToBo 1.8.13.1 and MiToBo-plugins 1.8.13.1, respectively. There is now much more editing functionality available, e.g., filling holes or drawing of free-hand lines.

March 2018

  • The LabelImageEditor has been released in MiToBo 1.8.8 and MiToBo-plugins 1.8.8, respectively.

Technical Details



The label editor relies on analyzing and manipulating the grayscale values of segmented regions in a given image.
In detail, the most important functions are implemented as follows:

  • deleting regions:
    the grayscale value at the mouse-click position is determined and all pixels in the image sharing this value are set to black
  • joining regions:
    upon clicking into the first region its grayscale value is internally stored;
    when the user clicks into the second region which is to be merged with the first one
    the grayscale value at the click position is extracted and all pixels with this value are relabeled with the formerly stored grayscale value of the first region
  • relabel image:
    the image is thresholded with a threshold of zero to get a binary image of background and labeled regions;
    then a sequential component labeling is run on the image which assigns consecutive grayscale labels to all connected components in the foreground of the image
  • adjust contrast:
    the image is relabeled, however, the target labels are equidistantly spread over just the upper two third of the available grayscale range;
    the underlying motivation for this is given by the observation that low grayscale values are hard to distinguish from the background, thus, brighter labels improve visibility
  • fix borders:
    this functions seeks to eliminate borders between regions with identical labels which may for example occur after joining two regions;
    for removing borders for each background pixel in the image the grayscale values in a 7x7 neighborhood of the pixel are analyzed,
    and if there is only a single grayscale value found and more than half of the analyzed neighborhood pixels share this value, the corresponding background pixel gets the value as well;
    the neighborhood is currently fixed to 7x7, thus, borders which are wider cannot be closed, and it has to be noted that sometimes the function has to be run several times to remove all pixels belonging to a single border