Overview

For the full run, this is an outline of the process:

• Program is supplied with the path to a folder that contains a file named input.tif and features.tif as well as any options
  • input.tif is the original image
  • features.tif is an image that contains Feature Marks as outlined below
  • A full list of options can be found using the flag --help, most options are outlined below as well
• Waypoints are used to find the perimeter of the cell
• Shape factors are calculated (area, convex hull area, perimeter, solidity, circularity, aspect ratio)
  • Solidity, circularity and aspect ratio are dimensionless
  • Perimeter is in px
  • Area and convex hull area are in px^2

Installation

Windows

The Python(x,y) distribution has all of the tools needed. Download it here: http://python-xy.github.io/downloads.html

If you don't do the full installation, make sure you go through the list and include:

• OpenCV
• Scipy
• Numpy
• scikit-image
• Matplotlib

Terms

• Sector - Piece of an image

• Cell

• Segment - Piece of a cell wall.

  These are not defined by waypoints (i.e. a segment can include and pass through a waypoint)

Feature Marking

Feature marking is a way of guiding the algorithm. You will add marks with specific colors to a copy of the image called features.tif (or features.png, etc). Each mark gives different guidance to the algorithms.

• Add waypoints to guide the algorithm finding the cell perimeter.
• Add regions to include segments for analysis that do not fall on the perimeter.
• Add redlines to prevent something from being included.

Waypoints

Waypoints mark a general path for computing the perimeter of the cell.

The edge finding algorithm works by finding the Minimum Cost Path between each waypoint. The cost matrix is calculated by inverting the intensity of each pixel in the image by subtracting it from 255 (each pixel is an 8 bit integer, 2^8 - 1 = 255 (range starts at 0 not 1)).

By finding the minimum cost path, it is effectively finding the shortest but brightest path between the two points.

Waypoints are added to the image as a single pixel colored pure blue (Hex: #0000FF, RGB: (0, 0, 255)). They should be added when the minimum cost path between two points won't necessarily be the best estimation of the cell's edge. You can add additional waypoints to shorten the distance between waypoints, making it more costly for the algorithm to go out of the way.

As an example, the minimum cost path in this figure does not follow the apparent cell edge.

By adding 3 waypoints you can redirect the path finding algorithm to include the convex region.

Note: It is important when adding these to pay attention to the behavior of your image editor of choice. For example, in Gimp the paint brush will create "soft" edges instead of just setting all of the pixels to the value you selected. In Gimp make sure to use the pencil tool.

Brush tool:

Pencil tool:

Example

Example files are located in the directory examples/waypoints. features.tif shows an example of marked waypoints.

To run this example:

python run.py --path-check ../examples/waypoints/

This will show the computed path based on the waypoints provided.

Workflow

The --path-check flag will only run the path finding between waypoints and display the result. This will allow you to quickly test your waypoints without waiting for the longer parts of the program to run.

Regions

Boxes that mark a region of interest - i.e. perpendicular region

Segments within regions will be analyzed separately

Redlines

Pure Red (#FF0000 or (255, 0, 0))

Will not be crossed when expanding a segment

Fully redlined regions will be ignored when doing feature searches

Void Seeds

To analyze the coverage on the image you will need to add marks to the image that show where voids will grow from. Any piece of the image not covered by cells should be marked using a pure magenta color denoted by hex code #FF00FF and RGB value (255, 0, 255).

The algorithm will k-means cluster the image and then grow each region adding any neighbor pixels that have an intensity that falls within the bottom two clusters/bins that have not already been assigned to a different void seed. Pixels will continue to be added until a less than 15% of the attempted additions succeed.

You can use the growth rate limit to control how far the void will bleed into neighboring regions. By putting two seeds close together you limit their growth since they will quickly no longer be able to add any pixels where they meet. If you would like a void to be likely to spread past obstacles (i.e. a noisy area on the image that should be included in the void), the best position for the seed is in the middle of the area. In the case of a region with ill-defined but existing boundaries, putting multiple seeds in close quarters will make them more susceptible to obstacles stopping the growth.

Note: Seeds are denoted by contiguous region, so if you make two separate marks but they are touching, that will still only be one region.

Example

Example files are located in the directory examples/coverage. features.tif shows an example of marked void seeds.

To run this example:

python run.py --coverage ../examples/coverage/

Options

Examples

Shape Factors