1. Identification of Leaves by Interior Shape and Texture
Veronica Dooly – A-BTech
Jennifer LeBlanc – West Wilkes High School
Introduce us

2. ProScope HR Hand-held microscope
Ability to change magnification by changing lenses
30x
100x
400x
Ability to save images or videos
Found it difficult to focus
Refocusing every leaf
Compounded by leaves curling
All magnify leaves could only have a section photographed

3. Magnification Strengths
30X
100X
400X

4. Magnification Images
EDGE
CENTER
STEM
30 x

5. Patterns Observed Voronoi Polygons
Visual observation to find patterns
Saw Voronoi – occur in nature

6. Patterns Observed Giraffe
Observed patterns that reminded of other things
In addition to giraffe saw cheetah and reptilian

7. Patterns Observed Transformations
Same shape repeats in some leaves
Rotated or dilated (different size)

8. Best Magnification for Visual Pattern from Images per Leaf
Leaf Type
Type 1
Type 2
Type 3
Type 4
Type 5
Type 6
Type 7
Type 10
Type 11
Type 12
Type 13
Leaf sample 1 2 3
Magnification
400
100 30
Not
“Not” – original photo was best

9. ImageJ Adjusting the color threshold
Attempt to make patterns more visible
ImageAdjust Color Threshold
Manually adjusted by dragging
Same leaf and magnification but optimum RGB varied greatly

10. ImageJ Adjusting the black/white threshold
Attempt to make patterns more visible
ImageAdjust B&W Threshold
Manually adjusted by dragging
Each leaf had different best range

11. OpenSURF Vector of 64 descriptors around each keypoint
SURF – (Speeded Up Robust Features)
Detect keypoints and descriptors on images
Invariant to size or color
K-means clustering
Groups the descriptors into N clusters based on their distance from the nearest centroid
100 cluster groups
Vector of 64 descriptors around each keypoint
Invariance overcame problems with magnification
Centroids – Voronoi polygons
Classifiers
The CrossVal function divides a single database into folds and used one partition of leaves to test identification using training sections
External Validation uses separate training and testing directories
External Validation

12. Key Points Some keypoints not on leaf
In pictures with multiple leaves keypoints everywhere
SURF creates a vector of 64 descriptors gleaned around each key point

13. K-means Clustering
Keypoints are clustered into points that are centroids of a voronoi polygon

14. Leaf with All Clusters Displayed

15. Cluster 38 on Common Hazel

16. Results of Automated Processing
Classifying Method
Database(s)
Number of Samples
Linear Discriminate
Performance
Diagonal Linear Discriminate
CrossVal
Train Leaves
8,416
61.7%
52.7%
Test Leaves
3,150
59.0%
51.8%
External Validation
Train & Test Leaves
11,566
24.5%
19.7%
First set was 389 leaves leaving 10 unique samples to test
Chose to use larger data base – with 118 unique samples
LeafClef database used (contest)
Discrepancy between CrossVal and external validation
Batches
Test images not in training images
Type of image
Further – Voronoi, other classifiers (i.e. quadratic), SURF with magnification
Scan
Photograph
Pseudoscan
Linear Discriminate Performance
23.9%
5.3%
31.4%
Diagonal Linear Discriminate Performance
18.4%
5.2%
22.2%

17. Acknowledgements Dr. Mitch Parry Dr. Rahman Tashakkori Clint Guin
Dr. Mary Beth Searcy
The Department of Computer Science at Appalachian State University
The National Science Foundation

18. Resources
Crisan, S.; Tarnovan, I.; Crisan, T., “Radiation optimization and image processing algorithms in the identification of hand vein patterns.” Computer Standards & Interfaces, vol. 32, no. 3, pp , March 2010, DOI: doi.org/ /j.csi
Henries, D.; Tashakkori, R., ‘Extraction of leaves from herbarium images,’ Electro/Information Technology (EIT), 2012 IEEE International Conference, vol. 1, no. 6, pp. 6-8, May DOI: /EIT
MathWorks. (n.d.). Kmeans [Online]. Available:
Raza, S.; Parry R.; Moffitt, R.; Young, A.; Wang, M., “An analysis of scale and rotation invariance in the bag-of-features method for hitopathological image classification.” MICCAI 2011 Lecture Notes in Computer Science, vol., 6893, pp Available: DOI: / _9.
Rebollo-Monedero, D.; Solé, M.; Forné, J., “A modification of the k-means method for quasi-unsupervised learning”, Knowledge-Based Systems, vol. 37, pp , Available:
Schmitt, D.; McCoy, N., “Object classification and localization using surf descriptors.” December Available:
Xia, Q., “The formation of a tree leaf.” ESAIM.Control, Optimisation and Calculus of Variations, vol. 13, no. 2, pp , Available: DOI: /cocv:
Xiaodong, Z.; Xiaojie W., “Leaf vein extraction based on gray-scale morphology”, IJIGSP, vol.2, no.2, pp.25-31, Available: