Object recognizing apparatus and method

• US 8,068,687 B2
• Filed: 10/31/2008
• Issued: 11/29/2011
• Est. Priority Date: 10/31/2007
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Kozakaya, Tatsuo, Nishiyama, Masashi, Yamaguchi, Osamu, Takeshima, Hidenori
• Primary Examiner(s): Tabatabai; Abolfazl
• Application Number:
  US12/262,735
• Publication Number:
  US 20090110303A1
• Time in Patent Office: 1,124 Days
• Field of Search: 382/103, 382/107, 382/118, 382/159, 382/165, 382/190, 382/224, 382/225, 382/255, 382/287, 382/307, 382/391, 348/94, 348/152, 348/154, 348/155, 348/169, 348/170, 348/208.1, 348/208.2, 348/208.14, 348/208.16, 348/352
• US Class Current: 382/255
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 5/73 Deblurring; Sharpening
• CPC Class Codes: G06V 10/20 Image preprocessing
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

An object is identified by detecting an object area image of an object to be recognized from a degraded image, converting the object area image to a frequency area, extracting a feature vector which indicates the amount of blur, comparing the feature vector and a classified plurality of blurred images, obtaining a cluster which is the most similar to the feature vector, selecting one point spread function corresponding to the similar cluster, restoring the object area image to the image before being blurred using the point spread function, and comparing the restored image and a target image.

Claims

1. An object recognizing apparatus comprising:

an area acquiring unit configured to detect object area images of an object to be recognized from a degraded image which is degraded by blur;

a feature vector generating unit configured to generate a feature vector by converting the object area images respectively to frequency areas and extracting a feature vector indicating the amount of blur;

a storage unit configured to generate a plurality of blurred images by applying a plurality of point spread functions stored in advance individually to a plurality of training images without blur stored in advance, group the plurality of blurred images generated by the plurality of point spread functions into clusters, and store the respective clusters therein in one-to-one correspondence with the respective point spread functions;

an estimating unit configured to compare the feature vector and the plurality of burred images belonging to the respective clusters by a pattern recognition, obtain a cluster which is most similar to the feature vector, and select one of the point spread functions which corresponds to the most similar cluster;

a restoring unit configured to restore the object area image into the image before blurred using the selected point spread function; and

an identifying unit configured to compare the restored image and the target image to identify the object.

2. The apparatus according to claim 1, wherein the feature vector generating unit calculates the feature vector by obtaining an amplitude value of the object area image by Fourier transform, applying a logarithmic transform to the amplitude value, and then performing down sampling and normalization of a norm.

3. The apparatus according to claim 1, wherein the storage unit stores subspaces formed from the plurality of blurred images belonging to the respective clusters on the cluster-to-cluster basis, and the estimating unit compares the subspaces for the respective clusters and the feature vector by a subspace method.

4. The apparatus according to claim 1, wherein the area acquiring unit acquires the object area images by detecting a feature point of the object area images from the degraded image and applying the feature point to an average object area shape.

5. An object recognizing method comprising:

acquiring an area by detecting object area images of an object to be recognized from a degraded image which is degraded by blur;

generating a feature vector by converting the object area images respectively to frequency areas and extracting a feature vector indicating the amount of blur;

storing clusters by generating a plurality of blurred images by applying a plurality of point spread functions stored in advance individually to a plurality of training images without blur stored in advance, grouping the plurality of blurred images generated by the plurality of point spread functions into clusters, and storing the respective clusters in one-to-one correspondence with the respective point spread functions;

estimating one of the point spread functions by comparing the feature vector and the plurality of burred images belonging to the respective clusters by a pattern recognition method, obtaining a cluster which is most similar to the feature vector, and selecting one of the point spread functions which corresponds to the most similar cluster;

restoring the object area image into the image before blurred using the selected point spread function; and

identifying the object by comparing the restored image and the target image to recognize the object.

6. The method according to claim 5, wherein the generating includes calculating the feature vector by obtaining an amplitude value of the object area image by Fourier transform, applying a logarithmic transform to the amplitude value, and then performing down sampling and normalization of a norm.

7. The method according to claim 5, wherein the storing includes storing subspaces formed from the plurality of blurred images belonging to the respective clusters on the cluster-to-cluster basis, and comparing the subspaces for the respective clusters and the feature vector by the subspace method.

8. The method according to claim 5, wherein the acquiring includes acquiring the object area images by detecting a feature point of the object area images from the degraded image and applying the feature point to an average object area shape.