Object Class Recognition
Abhitej John.BB.Tech 4/4
CS08765
Outline
Introduction
OCR’s role today
Algorithms in OCR
Neural Networks
Random forests
Conclusion
References
What is OCR?
A sub domain of Computer Vision
Related Domains – Pattern Recognition, Image Processing, Artificial Intelligence
Sub domains – object recognition, Video tracking,
Applications – Face detection, barcode decoding, Robotics
“Computer vision is the science and technology of machines that see”
What is Ocr? (continued..)
Steps involved
Segmentation (pixel color, intensity, texture..)
Clustering methods, Histogram based methods, edge detection, neural networks
Recognition
Geometry based, Appearance based and Feature based models.
Classification/ Detection
Neural Networks, Bayesian classifiers, Decision trees
Why OCR?
Applications.
Face/handwriting recognition systems
Autonomous vehicle navigation
Robotics
Gesture recognition
Content based image retrieval
Search by image
How OCR is done
Challenges
Object Variations – rotation, pose, scale
Occlusion
Illumination
Multiple objects
Learning an object model
How OCR is done (Continued..)
How OCR is Done (Continued..)
Classification of algorithms
Geometry based models
Appearance based models
Feature based models
Account for image variations
OCR with neural networks
Pre-processing step to extract shapes
Shape descriptors
Shape interpretation
Object representation
Object Interpretation
Classified object
OCR with Neural Networks
Shape interpretation
Object recognition Self Organizing Map
Object Classification Neural Network
OCR with neural networks
Training
Shape interpretation Neural Network
SOM for shape combinations
Classifier Neural Network
Recognition
OCR with Random forests
Random Forest
Classification
Features – multi class classifier, efficiency
Improvement on decision trees
Growing a tree If the number of cases in the training set is N, sample N cases at random - but with replacement, from
the original data. This sample will be the training set for growing the tree.
If there are M input variables, a number m<<M is specified such that at each node, m variables are selected at random out of the M and the best split on these m is used to split the node. The value of m is held constant during the forest growing.
Each tree is grown to the largest extent possible. There is no pruning.
OCR with random forests
Features
Node function
Abstest and difftest
OCR with Random foretss
Training and Testing
Choosing the training data
Tree creation based on information gain
Final outcome to test
OCR with random forests
A similar approach
Steps involved – segmentation, classification
Computing features
Node – appearance and shape
Conclusion
Other algorithms for recognizing object classes:
K-nearest neighbors and other clustering methods
SIFT
Decision trees
Bayesian classifiers
Unsupervised learning
Conclusion
Concepts covered
Computer Vision
Object Class Recognition
Neural Networks
Self Organizing Map
Random Forests
References
Florian Schroff, Antonio Criminisi, and Andrew Zisserman, Object Class Segmentation using Random Forests, in Proc. British Machine Vision Conference (BMVC), 2008
Object Recognition: A Shape-Based Approach using Artificial Neural Networks; Jelmer de Vries; MSc thesis, University of Utrecht Department of Computer Science, 2006
J. Winn and A. Criminisi, “Object Class Recognition at a Glance”. IEEE Computer Vision and Pattern Recognition (CVPR), New York, 2006.
Pinto N, Cox DD, DiCarlo JJ (2008) Why is Real-World Visual Object Recognition Hard? PLoS Comput Biol 4(1): e27. doi:10.1371/journal.pcbi.0040027.
J. Winn, A. Criminisi, and T. Minka. Object Categorization by Learned Universal Visual Dictionary. ICCV, 2005.
Thank You