190 likes | 201 Views
This article discusses the design and validation of classifiers for detecting and classifying errors in computer vision using machine learning. It explores the decision boundaries in complex data patterns and provides insights into the design of a classifier for specific applications. The article also covers the estimation of errors in a learning system and the importance of training and testing samples for accurate error estimation.
E N D
i= i=
Computer Vision Detection of Errors A/D Converter Sensor Object
+ + + + + o + + + + o + o o o o o o o o Pattern of Data X1 X2
Samples Learning System Classifiers Learning System
Classification Systems Data for classification Decision Pertaining to class Classifier
Samples for training Classifier for Specific application Case Variables (Features) Classes Learning System Classes Pertaining to samples Values of variables (xi) Classifier type Design of a classifier
Class (+) Class (-) Classification (+) Correct (+/+) Error (-/+) Classification (-) Error (+/-) Correct (-/-) Estimating the execution of a learning system What is an error? Reason for error (estimate) = number of errors number of cases
Apparent and true error Samples for training New cases Classifier Apparent reason for error True error
Samples Error estimationusing samples for training and samples for testing Cases for training the classifier Cases for testing the classifier
Example: 1-d Class 1: n1 = 5 X1 Train Y1 Train Class 2: n1 = 5 X2 Train Y2 Test
Classification ML Rule Class 1
Classification ML Rule Class 2
A simpler Classification ML Rule Class 1
Classification ML Rule Class 2