US 7,890,443 B2 · Learning classifiers using combined boosting and weight trimming

Abstract

A “Classifier Trainer” trains a combination classifier for detecting specific objects in signals (e.g., faces in images, words in speech, patterns in signals, etc.). In one embodiment “multiple instance pruning” (MIP) is introduced for training weak classifiers or “features” of the combination classifier. Specifically, a trained combination classifier and associated final threshold for setting false positive/negative operating points are combined with learned intermediate rejection thresholds to construct the combination classifier. Rejection thresholds are learned using a pruning process which ensures that objects detected by the original combination classifier are also detected by the combination classifier, thereby guaranteeing the same detection rate on the training set after pruning. The only parameter required throughout training is a target detection rate for the final cascade system. In additional embodiments, combination classifiers are trained using various combinations of weight trimming, bootstrapping, and a weak classifier termed a “fat stump” classifier.

Claims20

Dependent claims

A computer-readable medium having computer executable instructions stored thereon for automatically learning a set of weak classifiers of a combination classifier, said computer executable instructions comprising:

receiving a set of positive examples and a set of negative examples of instances in one or more sets of training data;
randomly sampling the set of negative examples to form a subset of negative examples;
initializing a weight for each example of the set of positive examples and the subset of negative examples;
for each of a predetermined number of weak classifiers, iteratively applying adaptive boosting-based learning to the set of positive examples and the subset of negative examples for a number of iterations to automatically learn a set of weak classifiers;
iteratively weight trimming examples from the set of negative examples and reforming the subset of negative examples while iteratively applying adaptive boosting-based learning;
and
outputting the learned set of weak classifiers.

4 more elements

A process for learning a set of weak classifiers, comprising:

receiving a set of positive examples of instances represented in one or more sets of training data;
receiving a set of negative examples of the instances;
selecting from the set of negative examples to form a subset of negative examples;
weighting each positive example;
weighting each negative example of the subset of negative examples;
automatically learning a predetermined number of weak classifiers by iteratively applying adaptive boosting-based learning to the set of positive examples and the subset of negative examples for a predetermined number of iterations;
iteratively removing a desired percentage of the lowest weighted negative examples from the set of negative examples and reforming the subset of negative examples along with new weights for the negative examples while iteratively applying the adaptive boosting-based learning;
and
outputting the predetermined number of learned weak classifiers.

6 more elements

Dependencies are inferred from the claim text.

Learning classifiers using combined boosting and weight trimming

Life of this patent

Learning classifiers using combined boosting and weight trimming

Life of this patent

Abstract

Claims20