Method, apparatus, and program for detecting objects in digital image

• US 20070036431A1
• Filed: 08/09/2006
• Published: 02/15/2007
• Est. Priority Date: 08/09/2005
• Status: Active Grant

• Current Assignee: Fujifilm Corporation (Fujifilm Holdings Corporation)
• Original Assignee: Fuji Photo Film Co Limited (Fujifilm Holdings Corporation)
• Inventors: Terakawa, Kensuke
• Granted Patent: US 7,760,940 B2
• Time in Patent Office: Days
• US Class Current: 382/170
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

In a method of detection of different objects in an input image by application, to partial images cut at different positions in the input image, of a plurality of weak classifiers that evaluate whether a detection target image is an image of a predetermined object based on a histogram of values of characteristic quantities calculated from a plurality of sample images representing the predetermined object, the histogram is extended to multi-dimensions and a criterion for the evaluation by the weak classifiers is a multi-dimensional histogram representing histograms for the different objects in the form of vectors.

Claims

1. A method of objects detection for detecting different objects in an input image, the method comprising the steps of:

preparing a plurality of weak classifiers each of which selects a vector in an N−

1 (N≧

3) dimensional histogram, the vector corresponding to values of characteristic quantities related to distribution of luminance calculated from each of partial images of a predetermined size cut from the input image, the N−

1 dimensional histogram generated by;

obtaining N histograms of values of the characteristic quantities for predetermined objects of N types through calculation of the characteristic quantities for each of the N types from different sample images representing the predetermined objects;

converting values of frequency in the respective N histograms into vectors by linearly combining the values of frequency with predetermined N basis vectors corresponding to the N histograms one to one, the basis vectors having the same magnitude and isotropy in an N−

1 dimensional space; and

obtaining the N−

1 dimensional histogram by combining the vectors of frequency for the values of the characteristic quantities corresponding to each other between the N histograms;

cutting the partial images of the predetermined size at different positions in the input image; and

carrying out judgment as to which one of the N types of the predetermined objects each of the partial images represents by;

combining the vector or vectors selected by at least one of the weak classifiers by applying the classifier or classifiers on each of the partial images;

extracting components of the basis vectors of the combined vector as scores respectively representing probabilities of the corresponding partial image being the predetermined objects of the N types corresponding to the basis vectors; and

carrying out the judgment based on magnitude of the scores.

2. The method of objects detection according to claim 1, wherein the weak classifiers are connected serially, and the step of carrying out the judgment is the step of judging that each of the partial images represents one of the predetermined objects by:

adding the vectors selected by the weak classifiers by sequentially applying the weak classifiers from the first classifier to each of the partial images; and

judging each of the partial images being an image of the predetermined object corresponding to one of the scores extracted according to the combined vector in the case where the score exceeds a predetermined threshold value.

3. The method of objects detection according to claim 2, wherein the weak classifiers have been selected from a plurality of weak classifiers according to a method of learning by boosting.

4. The method of objects detection according to claim 3, wherein the weak classifiers are connected in order determined according to the method of learning by boosting.

5. An apparatus of objects detection for detecting different objects in an input image, and the apparatus comprising:

a plurality of weak classifiers each of which selects a vector in an N−

1 (N≧

3) dimensional histogram, the vector corresponding to values of characteristic quantities related to distribution of luminance calculated from each of partial images of a predetermined size cut from the input image, the N−

1 dimensional histogram generated by;

obtaining N histograms of values of the characteristic quantities for predetermined objects of N types through calculation of the characteristic quantities for each of the N types from different sample images representing the predetermined objects;

converting values of frequency in the respective N histograms into vectors by linearly combining the values of frequency with predetermined N basis vectors corresponding to the N histograms one to one, the basis vectors having the same magnitude and isotropy in an N−

1 dimensional space; and

obtaining the (N−

1) dimensional histogram by combining the vectors of frequency for the values of the characteristic quantities corresponding to each other between the N histograms;

partial image cutting means for cutting the partial images of the predetermined size at different positions in the input image; and

judgment means for carrying out judgment as to which one of the N types of the predetermined objects each of the partial images represents by;

combining the vector or vectors selected by at least one of the weak classifiers by applying the classifier or classifiers on each of the partial images;

extracting components of the basis vectors of the combined vector as scores respectively representing probabilities of the corresponding partial image being the predetermined objects of the N types corresponding to the basis vectors; and

carrying out the judgment based on magnitude of the scores.

6. The apparatus of objects detection according to claim 5, wherein the weak classifiers are connected serially, and the judgment means judges that each of the partial images represents one of the predetermined objects by:

adding the vectors selected by the weak classifiers by sequentially applying the weak classifiers from the first classifier to each of the partial images; and

judging each of the partial images being an image of the predetermined object corresponding to one of the scores extracted according to the combined vector in the case where the score exceeds a predetermined threshold value.

7. The apparatus of objects detection according to claim 6, wherein the weak classifiers have been selected from a plurality of weak classifiers according to a method of learning by boosting.

8. The apparatus of objects detection according to claim 7, wherein the weak classifiers are connected in order determined according to the method of learning by boosting.

9. A program for causing a computer to function as means of objects detection for detecting different objects in an input image, the program causing the computer to function as:

a plurality of weak classifiers each of which selects a vector in an N−

1 (N≧

3) dimensional histogram, the vector corresponding to values of characteristic quantities related to distribution of luminance calculated from each of partial images of a predetermined size cut from the input image, the N−

1 dimensional histogram generated by;

obtaining N histograms of values of the characteristic quantities for predetermined objects of N types through calculation of the characteristic quantities for each of the N types from different sample images representing the predetermined objects;

converting values of frequency in the respective N histograms into vectors by linearly combining the values of frequency with predetermined N basis vectors corresponding to the N histograms one to one, the basis vectors having the same magnitude and isotropy in an N−

1 dimensional space; and

obtaining the (N−

1) dimensional histogram by combining the vectors of frequency for the values of the characteristic quantities corresponding to each other between the N histograms;

partial image cutting means for cutting the partial images of the predetermined size at different positions in the input image; and

judgment means for carrying out judgment as to which one of the N types of the predetermined objects each of the partial images represents by;

combining the vector or vectors selected by at least one of the weak classifiers by applying the classifier or classifiers on each of the partial images;

extracting components of the basis vectors of the combined vector as scores respectively representing probabilities of the corresponding partial image being the predetermined objects of the N types corresponding to the basis vectors; and

carrying out the judgment based on magnitude of the scores.

10. The program according to claim 9, wherein the weak classifiers are connected serially and the judgment means judges that each of the partial images represents one of the predetermined objects by:

adding the vectors selected by the weak classifiers by sequentially applying the weak classifiers from the first classifier to each of the partial images; and

judging each of the partial images being an image of the predetermined object corresponding to one of the scores extracted according to the combined vector in the case where the score exceeds a predetermined threshold value.

11. The program according to claim 10, wherein the weak classifiers have been selected from a plurality of weak classifiers according to a method of learning by boosting.

12. The program according to claim 11, wherein the weak classifiers are connected in order determined according to the method of learning by boosting.