Method for recognizing object using covariance descriptor and apparatus thereof

• US 9,418,429 B2
• Filed: 08/01/2014
• Issued: 08/16/2016
• Est. Priority Date: 08/02/2013
• Status: Expired due to Fees

• Current Assignee: Postech Academy-Industry Foundation
• Original Assignee: Postech Academy-Industry Foundation
• Inventors: Kim, Dai Jin, Yun, Jae Sik, Kim, In Su
• Primary Examiner(s): Koziol, Stephen R
• Assistant Examiner(s): Azima, Shaghayegh
• Application Number:
  US14/450,166
• Publication Number:
  US 20150036938A1
• Time in Patent Office: 746 Days
• Field of Search: 382/201
• US Class Current: 1/1
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 7/74 involving reference images or patches
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

Disclosed are object recognizing method and apparatus. The method may comprise extracting a feature point of an object to be recognized in an input image; generating a plurality of feature images by extracting orientation information of a region adjacent to the feature point and normalizing the region based in the orientation information; deriving a covariance descriptor based on the feature images; and recognizing the object in the input image by comparing the covariance descriptor with a reference covariance descriptor stored in advance. Thus, the object can be recognized fast and robustly to changes on the size of the input image, rotations of the input image, and illumination, whereby the recognition performance can be enhanced.

Claims

1. A method for recognizing objects, performed in an information processing apparatus which can perform digital signal processing, the method comprising:

extracting a feature point of an object to be recognized in an input image;

generating a plurality of feature images by extracting orientation information of a region adjacent to the feature point and normalizing the region based on the orientation information;

deriving a covariance descriptor based on the plurality of feature images; and

recognizing the object in the input image by comparing the covariance descriptor with at least one reference covariance descriptor stored in advance,

wherein extracting the feature point of the object to be recognized includes;

downscaling the input image into multiple downscaled images; and

extracting the feature point in corners of the object in the multiple downscaled images, and

wherein generating the plurality of feature images includes;

generating a feature point region by deriving orientation information of the region adjacent to the feature point;

generating a patch image by sampling values of pixels located within the feature point region and normalizing the feature point region; and

generating the plurality of feature images based on the patch image.

2. The method of claim 1, wherein deriving the covariance descriptor includes:

deriving a covariance matrix based on pixels located at same coordinate in the plurality of feature images; and

converting the covariance matrix into the covariance descriptor which is a vector in Euclidean space.

3. The method of claim 2, wherein the covariance matrix represents relationship between the pixels located at same coordinate in the plurality of feature images about characteristic including at least one of orientation of the object, color of the object, and brightness of the object.

4. The method of claim 1, wherein recognizing the object in the input image includes:

deriving an inlier matched pair by comparing the covariance descriptor with at least one reference covariance descriptor stored in advance; and

recognizing the object in the input image by estimating a homography of the object based on the inlier matched pair.

5. The method of claim 1, wherein the at least one reference covariance descriptor is derived from a plurality of reference images including the object, and stored in advance.

6. An apparatus for recognizing objects from an input image, which is equipped in an information processing apparatus capable of performing digital signal processing, the information processing apparatus comprising a digital signal processing (DSP) unit and a memory which is connected to the DSP unit, the apparatus performing modules stored in the memory, the modules comprising:

a feature point extracting part extracting a feature point of an object to be recognized in an input image;

a feature image generating part generating a plurality of feature images by extracting orientation information of a region adjacent to the feature point and normalizing the region based on the orientation information;

a descriptor deriving part deriving a covariance descriptor based on the feature images; and

an object recognizing part recognizing the object in the input image by comparing the covariance descriptor with a reference covariance descriptor stored in advance,

wherein the feature point extracting part downscales the input image into multiple downscaled images and extracts the feature point in corners of the object in the multiple downscaled images, and

wherein the feature image generating part includes;

a region generating module generating a feature point region by deriving orientation information of the region adjacent to the feature point; and

an image generating module generating a patch image by sampling values of pixels located within the feature point region and normalizing the feature point region and generating the plurality of feature images based on the patch image.

7. The apparatus of claim 6, wherein the descriptor deriving part generates the covariance descriptor by deriving a covariance matrix based on pixels located at same coordinate in the plurality of feature images and converting the covariance matrix into a vector in Euclidean space.

8. The apparatus of claim 7, wherein the covariance matrix represents relationship between the pixels located at same coordinate in the plurality of feature images about characteristic including at least one of orientation of the object, color of the object, and brightness of the object.

9. The apparatus of claim 6, wherein the object recognizing part derives an inlier matched pair by comparing the covariance descriptor with at least one reference covariance descriptor stored in advance, and recognizes the object in the input image by estimating a homography of the object based on the inlier matched pair.

10. The apparatus of claim 6, further comprising a descriptor data base storing the at least one reference covariance descriptor derived from a plurality of reference images including the object.