Estimating apparatus, method thereof, and computer program product therefor

1. An estimating apparatus comprising:

a memory that stores computer executable instructions; and

a processor, coupled to the memory, that executes the computer executable instructions to perform operations, comprising:

acquiring an image;

extracting human feature quantity from the image;

calculating, from the feature quantity, a first likelihood which indicates a degree of likelihood that the feature quantity belongs to for each of attribute classes, which comprises segments of consecutive attribute values relating to a person;

designating one of the attribute classes as a target class, designating two or more of the attribute classes near the target class as selected, and summing up the first likelihood of the target class and the first likelihoods of selected classes to obtain the second likelihood of the target class;

specifying one of the attribute classes as a specific attribute class, which has the highest second likelihood, from among the second likelihoods respectively for the attribute classes;

calculating an estimated attribute value of the specific attribute class and estimated attribute values of the selected classes by setting the specific attribute class as the target class, respectively by using the feature quantity; and

applying the second likelihood of the specific attribute class on the estimated attribute value of the specific attribute class as a weight to obtain a first value, applying the second likelihoods of the selected classes respectively on the estimated attribute values of the selected classes as weights to obtain a second value, and summing up the first value and the second value to obtain a corrected attribute value of the specific attribute class.

2. The apparatus according to claim 1 , further comprising

adapting a target-class weight for the first likelihood of the target class and selected-class weights for the first likelihoods of the selected classes

in a manner that the target-class weight is larger than any of the selected-class weight, and

applying the target-class weight with the first likelihood of the target class to obtain a third value and apply the selected-class weights on the first likelihoods of the selected classes to obtain a fourth value, and sum up the third value and the fourth value to obtain the second likelihood.

3. The apparatus according to claim 1 , further comprising selecting a predetermined number of selected classes.

4. The apparatus according to claim 1 , further comprising selecting the selected classes from the attribute classes near the target class in a manner that one having the highest value of the first likelihood is selected first, one having the second highest value of the first likelihood is selected second, and the attribute classes are selected by the values of the first likelihood in a descendent order as the selected classes.

5. The apparatus according to claim 1 , further comprising selecting the selected classes from among the attribute classes which may be within a predetermined range from the target class.

6. The apparatus according to claim 1 , further comprising selecting as the selected classes the attribute classes directly adjacent to the target class or the attribute classes that are distanced from the target class, by one to ten of the attribute classes.

7. The apparatus according to claim 1 , further comprising adapting the attribute values included in one of the attribute classes, which are also included in an adjacent one of the attribute classes.

8. The apparatus according to claim 1 , wherein the attribute value is an age or an angle of facing direction of the person.

9. An estimating method comprising:

acquiring an image;

extracting human feature quantity from the image;

calculating, from the feature quantity, a first likelihood which indicates a degree of likelihood that the feature quantity belongs to for each of attribute classes, which comprises segments of a consecutive attribute values relating to a person;

designating one of the attribute classes as a target class, designating two or more of the attribute classes near the target class as selected classes, and summing up the first likelihood of the target class and the first likelihoods of the selected classes to obtain the second likelihood of the target class;

specifying one of the attribute classes as a specific attribute class, which has the highest second likelihood, from among the second likelihoods respectively for the respective attribute classes;

calculating an estimated attribute value of the specific attribute class and estimated attribute values of the selected classes by setting the specific attribute class as the target class respectively by using the feature quantity; and

applying the second likelihood of the specific attribute class on the estimated attribute value of the specific attribute class as a weight to obtain a first value, applying the second likelihoods of the selected classes on the estimated attribute values of the selected classes as weights to obtain a second value, and sum up the first value and the second value to obtain a corrected attribute value of the specific attribute class.

10. A computer program product comprising a non-transitory computer-readable medium containing a program executed by a computer, the program causing the computer to execute:

an acquiring function configured to acquire an image including a person;

a feature-extracting function configured to extract human feature quantity from the image;

a first likelihood calculating function configured to calculate, from the feature quantity, a first likelihood which indicates a degree of likelihood that the feature quantity belongs to for each of attribute classes, which comprise segments of a continuous attribute value relating to the person;

a second likelihood calculating function configured to designate one of the attribute classes as a target class, designate two or more of the attribute classes near the target class as selected classes, and sum up the first likelihood of the target class and the first likelihoods of the selected classes to obtain the second likelihood of the target class;

a specifying function configured to specify one of the attribute classes as a specific attribute class, which has the highest second likelihood, from among the second likelihoods respectively for the attribute classes;

an attribute value calculating function configured to calculate an estimated attribute value of the specific attribute class and estimated attribute values of the selected classes setting the specific attribute class as the target class respectively by using the feature quantity; and

an integrating function configured to apply the second likelihood of the specific attribute class on the estimated attribute value of the specific attribute class as a weight to obtain a first value, apply the second likelihoods of the selected classes respectively on the estimated attribute values of the selected classes as weights to obtain a second value, and sum up the first value and the second value to obtain a corrected attribute value of the specific attribute class.