Method and apparatus for generating, updating classifier, detecting objects and image processing device

1. A method for generating a binary-class classifier for distinguishing an object with a background, comprising the following steps:

generating a one-class object classifier using at least one object sample;

selecting background samples from background image regions obtained from a sequence of images, using the one-class object classifier;

clustering the selected background samples into at least one set of background samples;

generating at least one one-class background classifier by generating a one-class background classifier for each set of background samples; and

generating the binary-class classifier by assembling the one-class object classifier and the at least one one-class background classifier into the binary-class classifier.

2. The method according to claim 1 , wherein, the step of selecting background samples comprising:

sliding a search window in at least one scale across each of the background image regions and obtaining the content in the background image region which is defined by the search window as a background sample;

sorting the background samples, based on the distances between the respective background samples and the one-class object classifier; and

selecting background samples based on the sorting results of the background samples.

3. The method according to claim 1 , further comprising the following steps:

with respect to each one-class classifier among the one-class object classifier and the at least one one-class background classifier, selecting samples from newly added samples as candidate updating samples of this one-class classifier; and

updating each of the one-class classifiers by using its corresponding candidate updating samples.

4. The method according to claim 3 , further comprising the following step:

generating at least one new one-class classifier by using the samples among the newly added samples which are not candidate updating samples of the respective one-class classifiers; and

adding the generated at least one new one-class classifier to the binary-class classifier.

5. The method according to claim 3 , further comprising the following step:

removing invalid one-class classifiers from the binary-class classifier.

6. The method according to claim 3 , wherein, with respect to each one-class classifier, selecting candidate updating samples for this one-class classifier from newly added samples, based on the distances between the respective newly added samples and the one-class classifier.

7. The method according to claim 1 , further comprising the step of

detecting an object in images by using the generated binary-class classifier.

8. The method according to claim 7 , wherein, the binary-class classifier is generated by at least one one-class background classifier by using a one-class object classifier and background image regions obtained from a sequence of images; and

assembling the one-class object classifier and the at least one one-class background classifier into a binary-class classifier.

9. The method according to claim 7 , further comprising the following steps:

collecting new object samples and background samples from a sequence of images so as to select samples from the new object samples and background samples as newly added samples; and

updating the binary-class classifier based on the newly added samples.

10. The method according to claim 9 , wherein, the binary-class classifier is updated, the binary-class classifier is generated by assembling a plurality of one-class classifiers which comprise a one-class object classifier and at least one one-class background classifier,

with respect to each one-class classifier among the plurality of one-class classifiers, selecting samples from newly added samples as candidate updating samples of this one-class classifier; and

updating each of the one-class classifiers by using its corresponding candidate updating samples.

11. The method according to claim 9 , wherein, the step of collecting new object samples and background samples from a sequence of images so as to select samples from the new object samples and background samples as newly added samples comprising:

sliding a search window in at least one scale across an image and obtaining the content in the image which is defined by the search window as a new sample;

sorting the new samples, based on the distances between the respective new samples and the one-class object classifier; and

selecting samples from these new samples as the newly added samples, based on the sorting results of these new samples.

12. An apparatus for generating a binary-class classifier for distinguishing an object with a background, the apparatus comprising:

a memory;

at least one processor coupled to the memory which executes the following:

generating a one-class object classifier using at least one object sample;

selecting background samples from background image regions obtained from a sequence of images, using the one-class object classifier;

clustering the selected background samples into at least one set of background samples;

generating at least one one-class background classifier by generating a one-class background classifier for each set of background samples; and

generating the binary-class classifier by assembling the one-class object classifier and the at least one one-class background classifier into the binary-class classifier.

13. The apparatus according to claim 12 , wherein, the background sample selecting component is further configured to:

slide a search window in at least one scale across each of the background image regions and obtain the content in the background image region which is defined by the search window as a background sample;

sort the background samples, based on the distances between the respective background samples and the one-class object classifier; and

select background samples based on the sorting results of the background samples.

14. The apparatus according to claim 12 , the apparatus further comprising:

selecting, with respect to each one-class classifier among the one-class object classifier and the at least one one-class background classifier, samples from newly added samples as candidate updating samples of this one-class classifier; and

updating each of the one-class classifiers by using its corresponding candidate updating samples.

15. The apparatus according to claim 14 , further comprising:

generating at least one new one-class classifier by using the samples among the newly added samples which are not candidate updating samples of the respective one-class classifiers; and

adding the at least one new one-class classifier to the binary-class classifier.

16. The apparatus according to claim 14 , further comprising:

removing invalid one-class classifiers from the binary-class classifier.

17. The apparatus according to claim 14 , wherein, candidate updating samples are selected for each one-class classifier from newly added samples, based on the distances between the respective newly added samples and the one-class classifier.

18. The apparatus according to claim 12 , further comprising:

detecting an object in images by using the generated binary-class classifier.

19. The apparatus according to claim 18 , wherein, the binary-class classifier is generated by using at least one one-class background classifier by using a one-class object classifier and background image regions obtained from a sequence of images; and

assembling the one-class object classifier and the at least one one-class background classifier into the binary-class classifier.

20. The apparatus according to claim 18 , further comprising:

collecting new object samples and background samples from a sequence of images so as to select samples from the new object samples and background samples as newly added samples; and

updating the binary-class classifier based on the newly added samples.

21. The apparatus according to claim 20 , wherein, the binary-class classifier is generated by assembling a plurality of one-class classifiers which comprise a one-class object classifier and at least one one-class background classifier,

with respect to each one-class classifier among the plurality of one-class classifiers, select samples from newly added samples as candidate updating samples of this one-class classifier; and

updating each of the one-class classifiers by using its corresponding candidate updating samples.

22. The apparatus according to claim 20 , wherein, the at least one processor is further configured to:

slide a search window in at least one scale across an image and obtain the content in the image which is defined by the search window as a new sample;

sort the new samples, based on the distances between the respective new samples and the one-class object classifier; and

select samples from these new samples as the newly added samples, based on the sorting results of these new samples.