Team D : Project #4
George Beretas – University College LondonDavid Papp - University of Pannonia
Gabor Retlaki - Pazmany Peter Catholic University
Ovidiu Adrian Turda - Technical University of Cluj-Napoca
Bark recognitionUsing Laws filters
For small texture: With 4 classes
For bigger texture like tree barks: With 6 classes
Common Hawthorn
Platanus × hispanica
Problems and possible solutions• These filters are not scale invariant, it is the cause of
bigger patches, and not a homogenous output image.• We could use Gabor filter to make the system scale
invariant.• Other possible solutions for recognition
– For feature extraction:• SIFT features• GLCM /gray level co-occurence matrix/
– For feature matching• Calculating cross correlation between features• Using mutual information
– For clustering• RANSAC• SVM• KNN
Leaf recognitionSegmentation of leaves - GrabCut
- GrabCut is an iterative image segmentation method based on graph cuts
- Needs user interaction
Hu moments- Hu moments are a set of image
moments- They are invariant under translation,
changes in scale, and rotation
Fourier moments- Calculate the distance between the
centroid and the boundary at certain angles
- Calculate DFT on this sequence
Classification
- Simple methods are used- Majority voting- k-nearest neighbors (with Euclidean
distance)
Problems and solutionsSmall data base
More samples
More test samples
Similarity between the testing and the data set leavesDifferent descriptorsMore complex classifiers
SummaryTree recognition based on leaves and barkBark recognition
Laws filterLeaf recognition
SegmentationFeature extractionClassification
Referenceshttps://code.ros.org/trac/opencv/browser/trunk/
opencv/samples/c/grabcut.cpp?rev=2326
http://en.wikipedia.org/wiki/Image_moment
http://en.wikipedia.org/wiki/K-nearest_neighbor_algorithm
Krishna Singh, Indra Gupta, Sangeeta Gupta, 2010, “SVM-BDT PNN and Fourier Moment Technique for
Classification of Leaf Shape”, International Journal of Signal Processing, Image Processing and Pattern Recognition Vol. 3, No. 4