Method, apparatus, and program for judging faces facing specific directions

Abstract

A partial image is cut out from an input image. A first classifier judges whether the partial image includes a face facing a direction within a first angular range. Another classifier judges whether the partial image includes a face facing a direction within another angular range that partially overlaps with the first angular range. The partial image is judged to include a face facing a direction within a specific angular range, which is the first angular range from which the portion that overlaps with the other angular range has been removed, if the first classifier judges that the image includes a face facing a direction within the first angular range, and the other classifier judges that the image does not include a face facing the second angular range. Each classifier is generated by machine learning, and judges whether images include faces facing directions corresponding to predetermined angular ranges.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram that illustrates the construction of a specific facing direction face detecting system 1, which is an embodiment of the present invention.

FIG. 2 is a diagram that illustrates an example of a conversion curve which is employed in a global normalizing process.

FIG. 3 is a diagram that illustrates the concept of a local region normalizing process.

FIG. 4 is a flow chart that illustrates the steps performed in the local region normalizing process.

FIG. 5 is a diagram that illustrates the detectable angular ranges of facing directions for each classifier of the present embodiment.

FIG. 6 is a block diagram that illustrates the construction of a classifier.

FIG. 7 is a flow chart that illustrates the processes performed by a single classifier.

FIG. 8 is a diagram for explaining how a weak classifier calculates characteristic amounts.

FIG. 9 is a flow chart that illustrates the learning method of a classifier.

FIG. 10 is a diagram that illustrates a sample image of a face, in which the positions of the eyes are normalized to predetermined positions.

FIG. 11 is a diagram that illustrates a method by which histograms of weak classifiers are generated.

FIG. 12 is a diagram that illustrates the concept behind generating a plurality of classifiers, each of which are capable of detecting faces facing directions within different angular ranges.

FIG. 13 is a diagram that illustrates how judgment is made regarding whether a face is facing a direction within a specific angular range, based on judgment results of classifiers.

FIG. 14 is a flow chart that illustrates the processes performed by the specific facing direction face detecting system 1.

FIGS. 15A, 15B and 15C are conceptual diagrams that illustrate the relationship between detection rates and the discriminatory nature of detection.

FIGS. 16A and 16B are diagrams that illustrate relationships between detectable angular ranges of facing directions for classifiers and the facing directions of faces which are actually detected.

Claims

1. A specific facing direction face detecting method, for detecting faces facing specific directions by employing classifiers that judge whether judgment target images include faces facing directions within angular ranges corresponding to predetermined angular ranges, generated by a machine learning technique employing a plurality of sample images that include faces having different shapes and facing directions within the predetermined angular ranges, comprising the steps of: cutting out a partial image from an input image;judging the partial image employing a first classifier that judges whether an image includes a face facing a direction within a first angular range;judging the partial image employing at least one other classifier that judges whether an image includes a face facing a direction within at least one other angular range, a portion of which overlaps with the first angular range; andjudging that the partial image includes a face facing a direction within a specific angular range, which is the first angular range from which the portion that overlaps with the at least one angular range corresponding to the at least one other classifier has been removed, in the case that partial image is judged to include a face facing a direction within the first angular range and judged not to include a face within the other angular range.

2. A specific direction facing face detecting method as defined in claim 1, wherein: the at least one other classifier is a second classifier for judging whether an image includes a face facing a direction within a second angular range that overlaps with one end of the first angular range; andthe specific angular range is the first angular range, from which the portion that overlaps with the second angular range has been removed.

3. A specific direction facing face detecting method as defined in claim 1, wherein: the at least one other classifier is a second classifier for judging whether an image includes a face facing a direction within a second angular range that overlaps with one end of the first angular range, and a third classifier for judging whether an image includes a face facing a direction within a third angular range that overlaps with the other end of the first angular range; andthe specific angular range is the first angular range, from which the portions that overlap with the second angular range and the third angular range have been removed.

4. A specific facing direction face detecting apparatus, for detecting faces facing specific directions by employing classifiers that judge whether judgment target images include faces facing directions within angular ranges corresponding to predetermined angular ranges, generated by a machine learning technique employing a plurality of sample images that include faces having different shapes and facing directions within the predetermined angular ranges, comprising: partial image cutout means, for cutting out a partial image from an input image;a first classifier, for judging whether the partial image includes a face facing a direction within a first angular range;at least one other classifier, for judging whether the partial image includes a face facing a direction within at least one other angular range, a portion of which overlaps with the first angular range; andjudging means, for judging that the partial image includes a face facing a direction within a specific angular range, which is the first angular range from which the portion that overlaps with the at least one angular range corresponding to the at least one other classifier has been removed, in the case that partial image is judged to include a face facing a direction within the first angular range and judged not to include a face within the other angular range.

5. A specific direction facing face detecting apparatus as defined in claim 4, wherein: the at least one other classifier is a second classifier for judging whether an image includes a face facing a direction within a second angular range that overlaps with one end of the first angular range; andthe specific angular range is the first angular range, from which the portion that overlaps with the second angular range has been removed.

6. A specific direction facing face detecting apparatus as defined in claim 4, wherein: the at least one other classifier is a second classifier for judging whether an image includes a face facing a direction within a second angular range that overlaps with one end of the first angular range, and a third classifier for judging whether an image includes a face facing a direction within a third angular range that overlaps with the other end of the first angular range; andthe specific angular range is the first angular range, from which the portions that overlap with the second angular range and the third angular range have been removed.

7. A program that causes a computer to function as a specific facing direction face detecting apparatus, for detecting faces facing specific directions by employing classifiers that judge whether judgment target images include faces facing directions within angular ranges corresponding to predetermined angular ranges, generated by a machine learning technique employing a plurality of sample images that include faces having different shapes and facing directions within the predetermined angular ranges, the program causing the computer to function as: partial image cutout means, for cutting out a partial image from an input image;a first classifier, for judging whether the partial image includes a face facing a direction within a first angular range;at least one other classifier, for judging whether the partial image includes a face facing a direction within at least one other angular range, a portion of which overlaps with the first angular range; andjudging means, for judging that the partial image includes a face facing a direction within a specific angular range, which is the first angular range from which the portion that overlaps with the at least one angular range corresponding to the at least one other classifier has been removed, in the case that partial image is judged to include a face facing a direction within the first angular range and judged not to include a face within the other angular range.

8. A program as defined in claim 7, wherein: the at least one other classifier is a second classifier for judging whether an image includes a face facing a direction within a second angular range that overlaps with one end of the first angular range; andthe specific angular range is the first angular range, from which the portion that overlaps with the second angular range has been removed.

9. A program as defined in claim 7, wherein: the at least one other classifier is a second classifier for judging whether an image includes a face facing a direction within a second angular range that overlaps with one end of the first angular range, and a third classifier for judging whether an image includes a face facing a direction within a third angular range that overlaps with the other end of the first angular range; andthe specific angular range is the first angular range, from which the portions that overlap with the second angular range and the third angular range have been removed.

10. A computer readable medium having the program defined in claim 7 recorded therein.

11. A computer readable medium having the program defined in claim 8 recorded therein.

12. A computer readable medium having the program defined in claim 9 recorded therein.