Image recognition apparatus with a plurality of classifiers

• US 20110026829A1
• Filed: 07/28/2010
• Published: 02/03/2011
• Est. Priority Date: 07/29/2009
• Status: Active Grant

• Current Assignee: DENSO Corporation
• Original Assignee: DENSO Corporation
• Inventors: Bandou, Takashi, Fukaya, Naoki
• Granted Patent: US 8,538,172 B2
• Time in Patent Office: Days
• US Class Current: 382/190
• CPC Class Codes: G06F 18/2148 characterised by the process organisation or structure, e.g. boosting cascade

Abstract

In an apparatus, an applying unit applies selected classifiers in sequence to an object image. A score calculating unit calculates, each time a classifier is applied to the object image, a summation of an output of an at least one classifier already applied to the object image to thereby obtain an acquisition score as the summation. The output of the at least one already applied classifier is weighed by a corresponding weight. A distribution calculating unit calculates, each time a classifier is applied to the object image, a predicted distribution of the acquisition score that would be obtained if at least one unapplied classifier in the classifiers, which has not yet been applied to the object image, were applied to the object image. A judging unit judges, based on the predicted distribution, whether to terminate an application of the at least one unapplied classifier to the object image.

Claims

1. An image recognition apparatus comprising:

a classifying unit comprising a plurality of classifiers, the plurality of classifiers being configured to be sensitive to images respectively having different specific patterns;

an applying unit configured to successively select the classifiers, and apply the selected classifiers in sequence to an input image as an object image;

a score calculating unit configured to calculate, each time one of the classifiers is applied to the object image by the applying unit, a summation of an output of at least one classifier already applied to the object image by the applying unit to thereby obtain an acquisition score as the summation, the output of the at least one already applied classifier being weighed by a corresponding weight;

a distribution calculating unit configured to calculate, each time one of the classifiers is applied to the object image by the applying unit, a predicted distribution of the acquisition score that would be obtained if at least one unapplied classifier in the classifiers, which has not yet been applied to the object image, were applied to the object image; and

a judging unit configured to judge, based on the predicted distribution calculated by the distribution calculating unit, whether to terminate an application of the at least one unapplied classifier to the object image by the applying unit.

2. An image recognition apparatus according to claim 1, wherein the judging unit comprises:

a predicted score and distribution range obtaining unit configured to obtain;

a predicted score by adding the acquisition score to an expectation value of the predicted distribution; and

a distribution range defined from an upper'"'"'limit to a lower limit based on a variance of the predicted distribution; and

a comparing unit configured to;

compare each of the upper limit and lower limit of the distribution range with a preset threshold; and

set the predicted score as a final acquisition score for the object image to terminate the application of the at least one unapplied classifier to the object image by the applying unit.

3. An image recognition apparatus according to claim 1, wherein the at least one unapplied classifier is provided in a plurality, the distribution calculating unit is configured to calculate an individual predicted distribution for each of the plurality of unapplied classifiers, and calculate at least one of an expectation value and a variance of the predicted distribution of the acquisition score by integrating the individual predicted distributions for the plurality of unapplied classifiers, and the applying unit is configured to select one of the plurality of unapplied classifiers to be next applied to the object image, the one of the plurality of unapplied classifiers having the highest variance in the plurality of unapplied classifiers.

4. An image recognition apparatus according to claim 1, wherein the object image is inputted in a plurality so that the applying unit is configured to successively select the classifiers, and apply the selected classifiers in sequence to a currently input object image in the plurality of object images, further comprising:

a storing unit configured to store therein information, the information including, for each of a plurality of previously set classes for classifying the plurality of object images, a first judgment probability of each of the classifiers making a positive judgment on at least one of the object images belonging to a corresponding one of the plurality of classes, and a second judgment probability of each of the classifiers making a positive judgment on at least one of the object images that does not belongs to a corresponding one of the plurality of classes; and

a probability calculating unit configured to calculate, for each of the plurality of classes, a posterior probability of the currently input object image belonging to a corresponding one of the plurality of classes,

the distribution calculating unit being configured to calculate the predicted distribution of the acquisition score based on the posterior probability and the information stored in the storing unit.

5. An image recognition apparatus according to claim 2, wherein the at least one unapplied classifier is provided in a plurality, and the predicted score and distribution range obtaining unit is configured to:

obtain an allowable value by multiplying a standard deviation of the predicted distribution by a factor of safety so determined as to be reduced with increase in a number of the plurality of unapplied classifiers; and

carry out at least one of an addition of the allowable value to the predicted score and a subtraction of the allowable value from the predicted score to thereby calculate the upper limit and lower limit of the distribution range defined.

6. An image recognition apparatus according to claim 5, wherein the factor of safety is normalized based on a number of the plurality of classifiers constituting the classifying unit.

7. An image recognition apparatus according to claim 1, wherein the object image is inputted in a plurality so that the applying unit is configured to successively select the classifiers, and apply the selected classifiers in sequence to a currently input object image in the plurality of object images, each of the plurality of object images is a sub-image extracted by a predetermined sized extraction window from an acquired image including at least one driver of a vehicle, and a plurality of previously set classes for classifying the plurality of object images includes a face class to which one or more object images including a predetermined facial region of the at least one driver belong, and a non-face class to which one or more object images that do not include the predetermined facial region of the at least one driver belong.

8. An image recognition method comprising:

providing a plurality of classifiers, the plurality of classifiers being configured to be sensitive to images respectively having different specific patterns;

successively selecting the classifiers;

applying the selected classifiers in sequence to an input image as an object image;

calculating, each time one of the classifiers is applied to the object image by the applying unit, a summation of an output of at least one classifier already applied to the object image by the applying to thereby obtain an acquisition score as the summation, the output of the at least one already applied classifier being weighed by a corresponding weight;

calculating, each time one of the classifiers is applied to the object image by the applying, a predicted distribution of the acquisition score that would be obtained if at least one unapplied classifier in the classifiers, which has not yet been applied to the object image, were applied to the object image; and

judging, based on the predicted distribution, whether to terminate an application of the at least one unapplied classifier to the object image by the applying.

9. An image recognition method according to claim 8, wherein the judging comprises:

obtaining;

a predicted score by adding the acquisition score to an expectation value of the predicted distribution; and

a distribution range defined from an upper limit to a lower limit based on a variance of the predicted distribution; and

comparing each of the upper limit and lower limit of the distribution range with a preset threshold; and

setting the predicted score as a final acquisition score for the object image to terminate the application of the at least one unapplied classifier to the object image by the applying.

10. A computer program product comprising:

a computer usable medium; and

a set of computer program instructions embodied on the computer useable medium, including instructions to;

successively select a plurality of classifiers, the plurality of classifiers being configured to be sensitive to images respectively having different specific patterns;

apply the selected classifiers in sequence to an input image as an object image;

calculate, each time one of the classifiers is applied to the object image by the applying unit, a summation of an output of at least one classifier already applied to the object image by the applying instruction to thereby obtain an acquisition score as the summation, the output of the at least one already applied classifier being weighed by a corresponding weight;

calculate, each time one of the classifiers is applied to the object image by the applying, a predicted distribution of the acquisition score that would be obtained if at least one unapplied classifier in the classifiers, which has not yet been applied to the object image, were applied to the object image; and

judge, based on the predicted distribution, whether to terminate an application of the at least one unapplied classifier to the object image by the applying instruction.

11. A computer program product according to claim 10, wherein the judging instruction comprises instructions to:

obtain;

a predicted score by adding the acquisition score to an expectation value of the predicted distribution; and

a distribution range defined from an upper limit to a lower limit based on a variance of the predicted distribution; and

compare each of the upper limit and lower limit of the distribution range with a preset threshold; and

set the predicted score as a final acquisition score for the object image to terminate the application of the at least one unapplied classifier to the object image by the applying instruction.