BARLEY SEEDS CLASSIFICATION IMAN SAUDY UMUT OGUR NORBERT
BARLEY SEEDS CLASSIFICATION IMAN SAUDY UMUT OGUR NORBERT KISS GEORGE TEPES-NICA
CONTENTS Introduction What is SVM? SVM Applications Text Categorization Face Detection The Approach About the Program Test results Conclusions
INTRODUCTION Barley seeds image Design a classifier Classes and statistical results
WHAT IS SVM? Linear algorithm in a high-dimensional space
WHAT IS SVM? A separable classification toy problem
WHAT IS SVM? Dot product Polynomial Kernel RBF Kernel Sigmoid Kernel
WHAT IS SVM? An Example Classifier Using RBF Kernel
ADVANTAGES Although it constructs models that are complex, it is simple enough to be analyzed mathematically It can lead to high performances in practical applications
SVM APPLICATIONS Text Categorization An Example – Reuters 12, 902 Reuters stories, 118 categories 75% to build classifiers 25% to test
SVM APPLICATIONS Face Detection MRI OCR
THE APPROACH Take several images for training (positive/negative) Tresholding to separate the seed from background Scale them and sub sample them to minimize the size of the vectors Feed them to the learning machine model/classifier
ABOUT THE PROGRAM Consists of two modules: for training for testing
TEST RESULTS training set: 28 p – 23 n errors: pos. images recognized as neg. images recognized as pos. training set: 43 p – 44 n errors: pos. images recognized as neg. images recognized as pos. 2 -4% 1 -2% 0% 0%
CONCLUSIONS SVMs are a good choice for binary classification (see results in this case) They can be used no matter what one may want to classify (faces, seeds, etc. ) For in-depth assistance join us for a beer tonight !!!
THANK YOU Team B
- Slides: 15