LemnaGrid

Image Processing without programming knowledge

LemnaGrid is the image analysis component developed by LemnaTec to address these specific needs in phenotyping. The image analysis is graphically represented as a process chain and can be subsequently saved as a LemnaTec image analysis Grid. Each Grid is constructed of Grid Devices, where a device is the graphical representation of an algorithm, e.g. loading an image from the database, applying a threshold etc.. Each Grid therefore starts with the Device to load the image and ends with the Device saving the result into the database. The construction of the Grid is performed via drag and drop of the Devices and connecting them to the chain. Once the chain is complete it can be used to automatically analyze all images of an experiment in the background.  

After acquiring the images it is crucial to extract all data for later statistical analysis. This task can be accomplished without knowledge about any programming languages. The 4 main steps to extract the parameters are, Load Images, find the object, calculate phonotypical properties and save them to the database. These steps are shown below.

Loading the example image

Load Reference Image

To construct a useful image processing grid it is mandatory to develop the grid with a set of images that represents all expected image characteristics. For example, if the goal is to apply a colour classification for green colour to the data, one should load images which show dark green plants as well as light green plants to ensure all sample colours are covered within the classification.
Consequently, the first step is selecting and loading those reference images that will encompass the full range of characteristics that the researcher may reasonably anticipate.

Finding the plant by the thresholding

Foreground/Background Seperation

The foreground/Background separation is the most important step of the grid as this step has the most influence on the quality of image analysis. The Grid provides multiple ways to accomplish that. Simple methods such as applying  a threshold give the researcher full control of the object detection or, by using more advanced computer vision algorithms like GrabCut, the researcher can accommodate outliers not represented in the reference images more effectifly than a user defined threshold.

Color classification of 5 green tones

Calculate Phenotypic Data

Beside the most common phenotypic parameters including digital biomass, height, width and colour classification, a Grid can also be constructed and applied to calculate more specific data such as the number of leaves. An often used techniques is the colour classification of images. Colors can be picked and their relative occurance can be quantified to detect effects like necrosis, lesions or to find fruits.

Save to Database

The final Device will be saving all the phenotypic parameters in the database. The saved information can then analyzed with another LemnaTec software module- LemnaMiner.
Before constructing a grid it is always wise to first check the  LemnaTec software module- LemanShare as other LemnaTec customers may have already posted a solution for public use or sharing.

Available Devices

Some of the avaiable Devices are listed below

Color spaces, use of different color space to minimize the effect of reflectance

  • RGB (standard representation of color)
  • HSI (Hue Saturation Value, closer to human perception)
  • LAB ( Luminance, a and b for the color opponent dimensions)
  • LUV (perceptual uniformity)

Edge Filters (Detect Edges in Images, such as outer border of the plant)

  • Laplace
  • Prewitt
  • Roberts
  • Scharr
  • Sobel
  • Canny Edge Detector
  • Various Filter
  • High/Low-pass filter (Enhance / reduce small structures in the image)
  • Median Filter
  • Gauss Filter

Morphological Operations

  • Closing        (useful to quantify the amount eaten by insects)
  • Opening      (can remove long small objects)
  • Dilatation    (inflates an object)
  • Erosion        (eliminates small objects)
  • Fill areas    (eliminates small objects and fills wholes in object at the  same time)
  • Skeleton    (Identify the structure of an object, Stem, leaves for maize)

Segmentation

  • Fore/Background Color     (find the desired object by selecting the color)
  • Nearest Neighbor    (find the desired object by selecting the color)
  • GrabCut (automatically seperates the plant based on an initial guess)
  • MeanShift  (enhances the color difference between object and background)

Threshold

  • Adaptive Threshold
  • Various

Un-distortion         (reduces the distortion effects of the lens and rectifies the image)

Movement Analysis    (Analyse the movement of small insects)

Classification

  • Color            (Determine necrotic areas in the plant)
  • Distance from point / line (classify symmetry)
  • Histogram (measure the complete color distribution in the plant)
  • Other properties    (Classify your plants for example by shape)