Systems and methods for training component-based object identification systems
AssigneeHonda Motor Co., Ltd.
InventorBernd Heisele
Life of this patent
Estimating the term…
Abstract
Systems and methods are presented that determine components to use as examples to train a component-based face recognition system. In one embodiment, an initial component shape and size is determined, a training set is built, a component recognition classifier is trained, and the accuracy of the classifier is estimated. The component is then temporarily grown in each of four directions (up, down, left, and right) and the effect on the classifier's accuracy is determined. The component is then grown in the direction that maximizes the classifier's accuracy. The process can be performed multiple times in order to maximize the classifier's accuracy.
Claims25
A method for determining a component, the component to be used by a component-based object identification system, the method comprising:
- using a processor to determine an accuracy of a first component recognition classifier, wherein the first component recognition classifier determines to which class a given component belongs and was trained based on a first component, the first component representing a first object region;
- determining a set of enlarged components, an enlarged component being larger than the first component, the enlarged component representing a second object region, wherein the second object region is larger than the first object region and contains the first object region;
- for each enlarged component in the set of enlarged components, determining an accuracy of a second component recognition classifier, wherein the second component recognition classifier determines to which class a particular component belongs and was trained based on the enlarged component;
- determining, from among the first component recognition classifier and each second component recognition classifier, which component recognition classifier is most accurate;
- and
- determining which component was used to train the most accurate component recognition classifier.
A system for determining a component, the component to be used by a component-based object identification system, the system comprising:
- means for determining an accuracy of a first component recognition classifier, wherein the first component recognition classifier determines to which class a given component belongs and was trained based on a first component, the first component representing a first object region;
- means for determining a set of enlarged components, an enlarged component being larger than the first component, the enlarged component representing a second object region, wherein the second object region is larger than the first object region and contains the first object region;
- means for determining, for each enlarged component in the set of enlarged components, an accuracy of a second component recognition classifier, wherein the second component recognition classifier determines to which class a particular component belongs and was trained based on the enlarged component;
- means for determining, from among the first component recognition classifier and each second component recognition classifier, which component recognition classifier is most accurate;
- and
- means for determining which component was used to train the most accurate component recognition classifier.
A computer program product for determining a component, the component to be used by a component-based object identification system, the computer program product comprising a computer-readable medium containing computer program code for performing a method, the method comprising:
- determining an accuracy of a first component recognition classifier, wherein the first component recognition classifier determines to which class a given component belongs and was trained based on a first component, the first component representing a first object region;
- determining a set of enlarged components, an enlarged component being larger than the first component, the enlarged component representing a second object region, wherein the second object region is larger than the first object region and contains the first object region;
- for each enlarged component in the set of enlarged components, determining an accuracy of a second component recognition classifier, wherein the second component recognition classifier determines to which class a particular component belongs and was trained based on the enlarged component;
- determining, from among the first component recognition classifier and each second component recognition classifier, which component recognition classifier is most accurate;
- and
- determining which component was used to train the most accurate component recognition classifier.
Dependencies are inferred from the claim text.