2DShapeDescription Code for performing a variety of 2D shape description using R

This code is specifically designed to work on binary images of strawberries where black pixels (value = 0) represent foreground (object) and white pixels (value = 1) are background (not object).

Please contact: mjfeldmann [at] ucdavis [dot] edu

1. Segmentation

   Lists the following: ImageJ Macro Description: Batch segments all images within a directory using two different segmentators
   File name: BatchSIOX_multiFruit.ijm
   Operating system(s): Platform independent
   Programming language: ImageJ

   We used imageJ to train two different SIOX segmentators using "BatchSIOX_multiFruit.ijm", referred to as "segmentator-NormalMulti.siox" and "segmentator-LightMulti.siox".
   We did this to account for the fact that strawberries with very light color were not properly segmented using a segmentator trained on your average red strawberries. Both generated images were averaged using "binary_image_mean.R" and pixel values = 0.5 were set to 0. Images were then checked for defects and holes in the segments. Holes were closed in ImageJ by hand using the 'Wand (tracing) tool".

2. Cropping, Edge Addition , and Scaling

   Lists the following: R code
   Description: Crops images based on positions of bounding box, adds white pixels, Scales all images to 1000x1000 pixels.
   File name: CropImages_ExtendBorders.R
   Operating system(s): Platform independent
   Programming language: R

   The bounding box of each object was extracted using ImageJ. This was done using the "brute force" method.
   Each binary multi fruit image is first opened in imageJ and all of the fruit in the image were individually selected with the "Wand (tracing) tool" and all of the measurements that imageJ performed were recorded. As this step was done by hand, there is no code in this repository for this task. Since all of the fruit land in 1 or 3 positions in all of the images, given a consisten orientation, we could have written an imageJ macro, similar to the segmentation macro to do this.

3. K-means Clustering

   Lists the following: R code
   Description: Flattens scaled binary images and performs K-means clustering.
   File name: Kmeans_Clustering.R
   Operating system(s): Platform independent
   Programming language: R

   Each image is first resized to 100x100px. The images are then flattened into 10,000 element vectors.
   Kmeans clustering is performed on this vector across all images.

4. Principal Progression of K Clusters

   Lists the following: R code Description: Takes the output from either Kmeans approach and returns a list containing the order of the factor levels.
   File name: PPKC.R
   Operating system(s): Platform independent
   Programming language: R

   All that goes in is a data.frame where each row is an image and each column is a value of K.
   Each cell is teh cluster assigned to a given image for a specific value of K.

5. Cumulative Logit Mixed Models

   Lists the following: R code
   Description: Estimates Heritability using a cumulative logit mixed model
   File name: VarComp_CLMM.R Operating system(s): Platform independent
   Programming language: R

   Used the 'ordinal::' R package to perform variance component estiamation using clmm().

6. Elliptical Fourier Analysis with Momocs

   Lists the following: R code
   Description: Performs EFA using the Momocs package in R
   File name: EFA_Momocs.R
   Operating system(s): Platform independent
   Programming language: R

   I had to specify thresholds for several images that wouldn't load properly and would crash R.

7. Procrustes analysis with Momocs

   Lists the following: R code
   Description: Performs GPA using the Momocs package in R
   File name: GPA_Momocs.R
   Operating system(s): Platform independent
   Programming language: R

   I had to specify thresholds for several images that wouldn't load properly and would crash R. This uses the same load sequence as EFA_Momocs.R

8. PCA of Psuedolandmarks

   Lists the following: R code
   Description: Performs PCA using the 50 landmarks used in GPA
   File name: Psuedolandmarks_PCA.R
   Operating system(s): Platform independent
   Programming language: R

   The first principal axis is extracted from each of the 50 landmarks. Those that are more variable than the median are kept for SEM.

9. Structured Equation Models of PsuedoLandmarks

   Lists the following: R code
   Description: Builds and evaluates a structured equation model using the Psuedolandmarks PCs.
   File name: SEM_Lavaan.R
   Operating system(s): Platform independent
   Programming language: R

10. Biomass Profile PCA

    Lists the following: R code
    Description: Calculated the biomass profile in the horizontal and vertical directions
    File name: BiomassProfile_PCA.R
    Operating system(s): Platform independent
    Programming language: R

    Binary images are scaled to 100x100. RowSums and ColSums are calculated for each of 100 rows and 100 columns. PCA is performed on the 100 element vectors of row or column sums.

11. EigenFruit Analysis

    Lists the following: R code
    Description: EigenFruit analysis is performed!
    File name: EigenFruit_PCA.R
    Operating system(s): Platform independent
    Programming language: R

    Binary images are scaled to 100x100 and flattened to 10,000 element vectors. PCA is performed on the 10,000 element vectors pixel values.

12. Variable Selection

    Lists the following: R code
    Description: Variable Selection using Random Forests is performed using VSURF.
    File name: VariablesSelection_Classification.R
    Operating system(s): Platform independent
    Programming language: R

    VSURF package is used to train random forests for variable selection.

13. Classification and Cross Validation

    Lists the following: R code
    Description: SVM and LDA are used to perfom shape classificaiton.
    File name: Validation_EXP.R
    Operating system(s): Platform independent
    Programming language: R

    A monumnetal for() loop that performs multiple instances of cross validation using SVM and LDA of the 9 values of K. Accuracy, Precision, FPR, and Recall are all calculated within the loop.