WO2009078701A2 - A multiple object symmetry axis extractor - Google Patents

A multiple object symmetry axis extractor Download PDF

Info

Publication number
WO2009078701A2
WO2009078701A2 PCT/MY2008/000181 MY2008000181W WO2009078701A2 WO 2009078701 A2 WO2009078701 A2 WO 2009078701A2 MY 2008000181 W MY2008000181 W MY 2008000181W WO 2009078701 A2 WO2009078701 A2 WO 2009078701A2
Authority
WO
WIPO (PCT)
Prior art keywords
symmetry axis
extraction
symmetry
axis
extractor
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/MY2008/000181
Other languages
French (fr)
Other versions
WO2009078701A3 (en
Inventor
Tomas Henrique Bode Mault
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Mimos Bhd
Original Assignee
Mimos Bhd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Mimos Bhd filed Critical Mimos Bhd
Publication of WO2009078701A2 publication Critical patent/WO2009078701A2/en
Publication of WO2009078701A3 publication Critical patent/WO2009078701A3/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/68—Analysis of geometric attributes of symmetry
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation

Definitions

  • the present invention relates to an object symmetry axis extractor and more particularly the present invention relates to a multiple object symmetry axis extractor.
  • both the said prior arts have an implicit assumption that axes are being extracted from single objects. However, in the present invention, multiple objects are explicitly taken account of. Further to this, the present invention, contrary to both prior art documents, computes centrality, width and elongation estimates of underlying objects. Further to this, the present invention uses Second Order Neural Networks wherein the present invention is not constrained by the use of edge orientation information (contrary to Fukushima). Further to that the present invention weighs axis votes based on factors such as feature rarity and salience.
  • the present invention relates to a multiple object symmetry axis extractor for symmetry axis extraction.
  • the present invention provides therein an extraction of symmetry axes from multiple objects in a 2D image and associates for each extracted axis an estimate of the underlying object's center, width and elongation along the axis.
  • the extraction is achieved via a second order neural network and wherein is further provided a voting structure consisting of a two-dimensional parameter space defined by ( ⁇ , r) parameters, wherein each pair of which specifies a different line corresponding to a particular symmetry axis.
  • the present invention After receiving a static image as input the present invention proceeds to the initialization of a voting structure which is characterized by the parameter space of interest such as angles and distances of symmetry axes, and estimates of centroids, widths and elongations of underlying objects. The present invention thereafter proceeds to compute a salience map of the inputted image.
  • the salience map is a representation of which image coordinates exhibit strong intensity variations and wherein the said salience map is used for sub-sampling the number of features that need to be considered when building second order representations. After computing the salience map, all pairs of salient points are compared, and those that are deemed to be sufficiently similar are selected for further processing.
  • each pair of matching salient points is used in order to compute a pair of symmetry axis parameters, a pair of centroid parameters and a width, all of which are used for voting.
  • the maxima for the symmetry axes parameters are found and wherein based on said maxima, the centroids and widths of the corresponding symmetry axes are averaged.
  • the elongation estimate for a particular symmetry axis is computed by averaging the distance between the centroids of all pairs of points contributing to that axis and the averaged centroid for that same axis.
  • Figure 1 shows a main process flow chart of an algorithm according to the present invention.
  • Figure 2 shows a comparison between a Second Order Neural Network on the left with a general Higher Order Neural Network on the right.
  • Figure 3 shows an illustration of four matching points (i.e. Pl, P2, P3 and P4) and the symmetry axes that originate from their paired coupling.
  • Figure 4 shows a simplified diagram of a second order neural network for the extraction of symmetry axes and the estimation of several geometric properties of underlying objects.
  • Figure 5 shows several examples of the invention's ability to extract symmetry axes.
  • Figure 6 shows a comparison of the present invention's ability to extract symmetry axes with that of a prior art, Fukushima et al. DETAILED DESCRIPTION OF THE PRESENT INVENTION
  • the invention extracts symmetry axes from multiple objects in a 2D image.
  • the invention also associates for each extracted axis an estimate of the underlying object's center, width and elongation along the axis.
  • the extraction is achieved via a second order neural network.
  • Figure 4 shows how node representing a symmetry axis is depicted in (a), while nodes (b), (c) and (d) represent width, centrality and elongation estimates respectively.
  • Second order nodes are represented at (g).
  • Several connection weights are depicted in (e).
  • the clouds in (f) represent small temporal delays.
  • the modulus of a difference is represented in (h) and divisive normalization is depicted at (i).
  • the voting structure consists of a two-dimensional parameter space defined by ( ⁇ , r) parameters, each pair of which specifies a different line corresponding to a particular symmetry axis.
  • FIG. 1 is summarized the main processes which characterize the functionality of the current invention.
  • a static image as input e.g. a grayscale image of height H and width W
  • the algorithm proceeds to initialize a voting structure which is characterized by the parameter space of interest, i.e.: angles and distances (relative to the origin) of symmetry axes, and estimates of centroids, widths and elongations of underlying objects.
  • the algorithm proceeds to compute a salience map of the inputted image.
  • a salience map is essentially a representation of which image coordinates exhibit strong intensity variations (e.g. edges).
  • the salience map is used for sub-sampling the number of features that need to be considered when building second order representations. After computing the above map, all pairs of salient points are compared, and those that are deemed to be sufficiently similar (e.g. in terms of their grayscale values) are selected for further processing,
  • each pair of matching salient points is used in order to compute a pair of symmetry axis parameters, a pair of centroid parameters and a width, all of which are used for voting.
  • the maxima for the symmetry axes parameters are found. Subsequently, based on these maxima, the centroids and widths of the corresponding symmetry axes are averaged.
  • the elongation estimate for a particular symmetry axis is computed by averaging the distance between the centroids of all pairs of points contributing to that axis and the averaged centroid for that same axis.
  • the symmetry axis, centroid and width parameters for a particular pair of matching salient points are computed as follows. To simplify the explanation we will refer to the points as A and B and the origin as O. First the bisector between A and B is computed. The bisector is the symmetry axis, but its parameters will be defined as the length and angle of a vector starting at the origin and intersecting (and ending at) the bisector perpendicularly. The centroid is defined as the midpoint between A and B on the line AB. The width is defined as the distance between A and B. Refer to Figure 3 for an illustration of the symmetry axes computed for 4 points.
  • Figure 5 in (a) demonstrates the ability to extract axes from multiple objects.
  • the subfigure in (b) demonstrates the invention's ability to extract the main symmetry axis of a face and the subfigure in (c) demonstrates the invention's ability to extract a further symmetry axis which is relevant to eye position.
  • Figure 6 illustrates a comparison of the proposed invention's ability to extract symmetry axes with that of Fukushima et al.
  • the invention's axes are incorporated with the patterns while those of Fukushima et al are in squares below each pattern.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Geometry (AREA)
  • Image Analysis (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The present invention relates to a multiple object symmetry axis extractor for symmetry axis extraction characterized in that wherein is provided the extraction of symmetry axes from multiple objects in a 2D image which associates for each extracted axis an estimate of the underlying object's center, width and elongation along the axis and wherein the extraction is achieved via a second order neural network and wherein is further provided a voting structure consisting of a two-dimensional parameter space defined by (α, r) parameters, wherein each pair of which specifies a different line corresponding to a particular symmetry axis.

Description

A MULTIPLE OBJECT SYMMETRY AXIS EXTRACTOR
FIELD OF INVENTION
The present invention relates to an object symmetry axis extractor and more particularly the present invention relates to a multiple object symmetry axis extractor.
BACKGROUND OF THE INVENTION
The closest patented invention is that of Saito US Patent No 5,809,170 wherein candidate symmetry axes in a target pattern are tested sequentially until one with sufficient evidence is found (or not) and then used in order to register the said pattern. Apart from Saito no other patents were found pertaining to symmetry axis extraction.
The closest academic system is that of Fukushima et al. wherein axes of symmetry are extracted via a hierarchical neural network with preprocessing layers performing contrast and edge extraction.
Both the said prior arts have an implicit assumption that axes are being extracted from single objects. However, in the present invention, multiple objects are explicitly taken account of. Further to this, the present invention, contrary to both prior art documents, computes centrality, width and elongation estimates of underlying objects. Further to this, the present invention uses Second Order Neural Networks wherein the present invention is not constrained by the use of edge orientation information (contrary to Fukushima). Further to that the present invention weighs axis votes based on factors such as feature rarity and salience.
Therefore, it is an objective of the invention to allow symmetry axis extraction from multiple objects in the same image. It is another objective of the present invention to form estimates of the centers, widths and elongations of underlying objects.
SUMMARY OF THE INVENTION
The present invention relates to a multiple object symmetry axis extractor for symmetry axis extraction. The present invention provides therein an extraction of symmetry axes from multiple objects in a 2D image and associates for each extracted axis an estimate of the underlying object's center, width and elongation along the axis. The extraction is achieved via a second order neural network and wherein is further provided a voting structure consisting of a two-dimensional parameter space defined by (α, r) parameters, wherein each pair of which specifies a different line corresponding to a particular symmetry axis.
After receiving a static image as input the present invention proceeds to the initialization of a voting structure which is characterized by the parameter space of interest such as angles and distances of symmetry axes, and estimates of centroids, widths and elongations of underlying objects. The present invention thereafter proceeds to compute a salience map of the inputted image.
The salience map is a representation of which image coordinates exhibit strong intensity variations and wherein the said salience map is used for sub-sampling the number of features that need to be considered when building second order representations. After computing the salience map, all pairs of salient points are compared, and those that are deemed to be sufficiently similar are selected for further processing.
Then each pair of matching salient points is used in order to compute a pair of symmetry axis parameters, a pair of centroid parameters and a width, all of which are used for voting. After all pairs of matching salient points have been considered, the maxima for the symmetry axes parameters are found and wherein based on said maxima, the centroids and widths of the corresponding symmetry axes are averaged. The elongation estimate for a particular symmetry axis is computed by averaging the distance between the centroids of all pairs of points contributing to that axis and the averaged centroid for that same axis.
BRIEF DESCRIPTION OF THE FIGURES
Figure 1 shows a main process flow chart of an algorithm according to the present invention.
Figure 2 shows a comparison between a Second Order Neural Network on the left with a general Higher Order Neural Network on the right.
Figure 3 shows an illustration of four matching points (i.e. Pl, P2, P3 and P4) and the symmetry axes that originate from their paired coupling.
Figure 4 shows a simplified diagram of a second order neural network for the extraction of symmetry axes and the estimation of several geometric properties of underlying objects.
Figure 5 shows several examples of the invention's ability to extract symmetry axes.
Figure 6 shows a comparison of the present invention's ability to extract symmetry axes with that of a prior art, Fukushima et al. DETAILED DESCRIPTION OF THE PRESENT INVENTION
The present invention will now be described in detail with reference made to the accompanied drawings but the scope of the invention not limited thereto.
The invention extracts symmetry axes from multiple objects in a 2D image. The invention also associates for each extracted axis an estimate of the underlying object's center, width and elongation along the axis. The extraction is achieved via a second order neural network. Refer to Figure 2 and Figure 4 for a diagrammatic and simplified depiction of the network, which compares pairs of features and votes for a particular axis contingent on a sufficient match being found for each pair. Figure 4 shows how node representing a symmetry axis is depicted in (a), while nodes (b), (c) and (d) represent width, centrality and elongation estimates respectively. Second order nodes are represented at (g). Several connection weights are depicted in (e). The clouds in (f) represent small temporal delays. The modulus of a difference is represented in (h) and divisive normalization is depicted at (i).
The voting structure consists of a two-dimensional parameter space defined by (α, r) parameters, each pair of which specifies a different line corresponding to a particular symmetry axis.
Reference is now made to Figure 1 wherein is summarized the main processes which characterize the functionality of the current invention. After receiving a static image as input (e.g. a grayscale image of height H and width W), the algorithm proceeds to initialize a voting structure which is characterized by the parameter space of interest, i.e.: angles and distances (relative to the origin) of symmetry axes, and estimates of centroids, widths and elongations of underlying objects. Following this, the algorithm proceeds to compute a salience map of the inputted image. A salience map is essentially a representation of which image coordinates exhibit strong intensity variations (e.g. edges).
The salience map is used for sub-sampling the number of features that need to be considered when building second order representations. After computing the above map, all pairs of salient points are compared, and those that are deemed to be sufficiently similar (e.g. in terms of their grayscale values) are selected for further processing,
After this, each pair of matching salient points is used in order to compute a pair of symmetry axis parameters, a pair of centroid parameters and a width, all of which are used for voting. After all pairs of matching salient points have been considered, the maxima for the symmetry axes parameters are found. Subsequently, based on these maxima, the centroids and widths of the corresponding symmetry axes are averaged.
Finally, the elongation estimate for a particular symmetry axis is computed by averaging the distance between the centroids of all pairs of points contributing to that axis and the averaged centroid for that same axis. The symmetry axis, centroid and width parameters for a particular pair of matching salient points are computed as follows. To simplify the explanation we will refer to the points as A and B and the origin as O. First the bisector between A and B is computed. The bisector is the symmetry axis, but its parameters will be defined as the length and angle of a vector starting at the origin and intersecting (and ending at) the bisector perpendicularly. The centroid is defined as the midpoint between A and B on the line AB. The width is defined as the distance between A and B. Refer to Figure 3 for an illustration of the symmetry axes computed for 4 points.
Reference could also be made to Figures 5 and 6 for comparison purposes.
Figure 5 in (a) demonstrates the ability to extract axes from multiple objects. The subfigure in (b) demonstrates the invention's ability to extract the main symmetry axis of a face and the subfigure in (c) demonstrates the invention's ability to extract a further symmetry axis which is relevant to eye position. Figure 6 illustrates a comparison of the proposed invention's ability to extract symmetry axes with that of Fukushima et al. The invention's axes are incorporated with the patterns while those of Fukushima et al are in squares below each pattern.

Claims

1. A multiple object symmetry axis extractor for symmetry axis extraction characterized in that wherein is provided the extraction of symmetry axes from multiple objects in a 2D image which associates for each extracted axis an estimate of the underlying object's center, width and elongation along the axis and wherein the extraction is achieved via a second order neural network and wherein is further provided a voting structure consisting of a two-dimensional parameter space defined by (α, r) parameters, wherein each pair of which specifies a different line corresponding to a particular symmetry axis.
2. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 1 wherein after receiving a static image as input the present invention would proceed to initialize a voting structure which is characterized by the parameter space of interest such as angles and distances of symmetry axes, and estimates of centroids, widths and elongations of underlying objects.
3. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 2 wherein the present invention would thereafter proceed to compute a salience map of the inputted image.
4. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 3 wherein the salience map is a representation of which image coordinates exhibit strong intensity variations and wherein the said salience map is used for sub- sampling the number of features that need to be considered when building second order representations.
5. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 4 wherein after computing the salience map, all pairs of salient points are compared, and those that are deemed to be sufficiently similar are selected for further processing.
6. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 3 wherein then each pair of matching salient points is used in order to compute a pair of symmetry axis parameters, a pair of centroid parameters and a width, all of which are used for voting.
7. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 3 wherein after ail pairs of matching salient points have been considered, the maxima for the symmetry axes parameters are found and wherein based on said maxima, the centroids and widths of the corresponding symmetry axes are averaged.
8. A multiple object symmetry axis extractor for symmetry axis extraction as claimed in Claim 1 wherein the elongation estimate for a particular symmetry axis is computed by averaging the distance between the centroids of all pairs of points contributing to that axis and the averaged centroid for that same axis.
PCT/MY2008/000181 2007-12-14 2008-12-12 A multiple object symmetry axis extractor Ceased WO2009078701A2 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
MYPI20072251 2007-12-14
MYPI20072251A MY176263A (en) 2007-12-14 2007-12-14 A method of extracting symmetry axes from multiple objects in 2d images

Publications (2)

Publication Number Publication Date
WO2009078701A2 true WO2009078701A2 (en) 2009-06-25
WO2009078701A3 WO2009078701A3 (en) 2009-10-15

Family

ID=40796044

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/MY2008/000181 Ceased WO2009078701A2 (en) 2007-12-14 2008-12-12 A multiple object symmetry axis extractor

Country Status (2)

Country Link
MY (1) MY176263A (en)
WO (1) WO2009078701A2 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2421247A1 (en) * 2010-08-19 2012-02-22 Sony Corporation Image processing device, method, and program
WO2018119785A1 (en) * 2016-12-28 2018-07-05 Intel Corporation Method and apparatus for a binary neural network mapping scheme utilizing a gate array architecture

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SG91841A1 (en) * 1999-11-03 2002-10-15 Kent Ridge Digital Labs Face direction estimation using a single gray-level image
JP4647289B2 (en) * 2004-11-10 2011-03-09 富士フイルム株式会社 Image processing method, apparatus, and program

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2421247A1 (en) * 2010-08-19 2012-02-22 Sony Corporation Image processing device, method, and program
US8737768B2 (en) 2010-08-19 2014-05-27 Sony Corporation Image processing device, method, and program
WO2018119785A1 (en) * 2016-12-28 2018-07-05 Intel Corporation Method and apparatus for a binary neural network mapping scheme utilizing a gate array architecture
US11055613B2 (en) 2016-12-28 2021-07-06 Intel Corporation Method and apparatus for a binary neural network mapping scheme utilizing a gate array architecture

Also Published As

Publication number Publication date
WO2009078701A3 (en) 2009-10-15
MY176263A (en) 2020-07-24

Similar Documents

Publication Publication Date Title
CN102880877B (en) Target identification method based on contour features
Campo et al. Multimodal stereo vision system: 3D data extraction and algorithm evaluation
Pears et al. Ground plane segmentation for mobile robot visual navigation
CN107958466B (en) Slam algorithm optimization model-based tracking method
CN101216895A (en) An Automatic Extraction Method of Ellipse Image Feature in Complicated Background Image
CN103778626A (en) Quick image registration method based on visual remarkable area
CN112487867A (en) Visual constraint fingerprint identification method based on enhanced triangulation
WO2009078701A2 (en) A multiple object symmetry axis extractor
CN104252701B (en) Correct the method and system of disparity map
CN105678720A (en) Image matching judging method and image matching judging device for panoramic stitching
CN107292238A (en) Contour of building recognition methods with end point is constrained based on spatial topotaxy
Liu et al. Keypoint matching by outlier pruning with consensus constraint
Zhao et al. An approach based on mean shift and kalman filter for target tracking under occlusion
Xiao et al. A topological approach for segmenting human body shape
CN107229899A (en) A kind of finger vein identification method based on dimensional Gaussian maximum curvature
CN102880869B (en) Based on the fingerprint direction field under Morkov random field condition extracting method of priori
Huang et al. Gait recognition using multiple views
CN109255150B (en) A data association method for multi-antenna angle of arrival based on bidirectional order association
Palraj et al. Retinal fundus image registration via blood vessel extraction using binary particle swarm optimization
Kim Geometry of vanishing points and its application to external calibration and realtime pose estimation
CN110210341A (en) Identity card authentication method based on face recognition, system thereof and readable storage medium
Luo et al. Matching Point-sets using Procrustes Alignment and the EM Algorithm.
Lebègue et al. Detecting 3-D parallel lines for perceptual organization
CN114821685A (en) PSO (particle swarm optimization) based non-contact 3D fingerprint identification method for optimizing Delaunay triangulation
Saval-Calvo et al. A comparative study of downsampling techniques for non-rigid point set registration using color

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 08861783

Country of ref document: EP

Kind code of ref document: A2

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 08861783

Country of ref document: EP

Kind code of ref document: A2