Method for recognition between a first object and a second object each represented by images

• US 8,027,532 B2
• Filed: 03/17/2006
• Issued: 09/27/2011
• Est. Priority Date: 03/24/2005
• Status: Expired due to Fees

• Current Assignee: Kee Square S.r.l. (CGE Investimenti SRL)
• Original Assignee: Kee Square S.r.l. (CGE Investimenti SRL)
• Inventors: Onofrio, Davide, Marcon, Marco
• Primary Examiner(s): LEE, JOHN W
• Application Number:
  US11/887,776
• Publication Number:
  US 20090136095A1
• Time in Patent Office: 2,020 Days
• Field of Search: None
• US Class Current: 382/156
• CPC Class Codes: G06F 18/2414 Smoothing the distance, e.g. radial basis function networks [RBFN]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A method for recognition between a first and second object represented by at least one first image and at least one second image, includes defining rectangular assemblies of random pixels; filtering the first image with first n filters obtained from the assemblies of pixels to obtain n first filtered matrices; classifying the n first filtered matrices by providing a first center and a first radius within a space of N dimensions; filtering the second image with the first n filters to obtain n second filtered matrices; classifying the n second filtered matrices by providing a second center within the space of N dimensions; and comparing the first center and first radius with the second center.

Claims

1. A method for recognition between a first object and a second object represented by at least one first image and at least one second image, each containing a plurality of pixels, comprising the following steps:

defining a plurality of randomly positioned rectangular array of pixels;

filtering by the mathematical operation of bi-dimensional matrix convolution of said first image with the first n matrices obtained from said randomly positioned rectangular array of pixels to obtain n first filtered matrices;

classifying said n first filtered matrices by providing a first centre and a first radius within a space of N dimensions, for each of said n first filtered matrices, said first radius and said first centre representing n first hyperspheres within a space of N dimensions;

filtering by the mathematical operation of bi-dimensional matrix convolution of said second image with the first n matrices obtained from said randomly positioned rectangular array of pixels to obtain n second filtered matrices;

classifying said n second filtered matrices by providing a second centre within said space of N dimensions, for each of said n second filtered matrices;

comparing said first centre and said first radius of each of said first filtered matrices with said second centre for each of said n second filtered matrices; and

considering that recognition between said first object and said second object has taken place if at least one said second centre lies at a distance from said first centre which is less than or equal to at least one said first radius,

wherein an inserted image of an unknown object is recognized only if a vector of the unknown object pertains to an N-dimensional sphere defined by the centre and radius of a previously inserted image.

2. The method as claimed in claim 1, further comprising:

inserting a third object represented by at least one third image;

providing a third radius within said space of N dimensions; and

dimensioning said third radius such that it does not interfere with said first radius.

3. The method as claimed in claim 1, wherein the step of classifying said n first and second filtered matrices takes place by using a RBF (radial basis function) neural network.

4. The method as claimed in claim 3, wherein said RBF (radial basis function) neural network comprises only one input layer and only one output layer.

5. The method as claimed in claim 1, further comprising the step of rescaling said first and second filtered matrices, by reducing their size by means of a bilinear interpolation.

6. The method as claimed in claim 1, wherein said objects are faces of individuals.

7. The method as claimed in claim 1, wherein said phases of classifying are carry out by decreasing values of the relative eigenvalues derived from the correlation matrix of said rectangular pixel assemblies.

8. The method as claimed in claim 1, further comprising the step of classifying said n second filtered matrices by providing a second radius and a second centre within said space of N dimensions, for each of said n second filtered matrices;

said first radius and said first centre representing n first hyperspheres within a space of N dimensions, said second radius and said second centre representing n second hyperspheres within a space of N dimensions, recognition between said first object and said second object being considered to have taken place if there is interference between at least one of said first hyperspheres and at least one of said second hyperspheres.

9. The method as claimed in claim 1, wherein the step of classifying said n first and second filtered matrices takes place by using six RBF (radial basis function) neural networks, and if the vectors obtained fall outside spheres of all neurons, the object is identified as not recognized.

10. The method as claimed in claim 1, wherein in one or more of the neural networks the vectors fall within neurons indicating a same category identifying the same object, whereas in other neural networks the vectors fall outside a radius of influence of all the neurons, whereby corresponding object is identified as that indicated by the category of the neurons which have recognized the corresponding object.

11. The method as claimed in claim 1, wherein if different neural networks give conflicting indications about the object to whom the photograph pertains, a weighted sum of the indications from each network is evaluated.

12. The method as claimed in claim 1, wherein if different neural networks give conflicting indications about the object to whom the photograph pertains, only those neural networks in which the vector obtained falls within a sphere of influence will be considered.

13. A method for recognition between a first object and a second object represented by at least one first image and at least one second image, each containing a plurality of pixels, comprising the following steps:

defining a plurality of randomly positioned rectangular array of pixels;

filtering by the mathematical operation of bi-dimensional matrix convolution of said first image with the first n matrices obtained from said randomly positioned rectangular array of pixels to obtain n first filtered matrices;

classifying said n first filtered matrices by providing a first centre and a first radius within a space of N dimensions, for each of said n first filtered matrices, said first radius and said first centre representing n first hyperspheres within a space of N dimensions;

filtering by the mathematical operation of bi-dimensional matrix convolution of said second image with the first n matrices obtained from said randomly positioned rectangular array of pixels to obtain n second filtered matrices;

classifying said n second filtered matrices by providing a second centre within said space of N dimensions, for each of said n second filtered matrices;

comparing said first centre and said first radius of each of said first filtered matrices with said second centre for each of said n second filtered matrices; and

considering that recognition between said first object and said second object has taken place if at least one said second centre lies at a distance from said first centre which is less than or equal to at least one said first radius,

wherein the steps of classifying said n first and second filtered matrices takes place by using a RBF (radial basis function) neural network, and a distance of a vector from a centre of a closest neuron of said neural network is evaluated by rule L1;

14. The method according to claim 13, wherein the distance of the vector from the centre of the closest neuron is evaluated such that each eigenpaxel is given a particular eigenvalue which is an index of variance possessed by the analyzed data in a direction indicated by the eigenpaxel.

15. The method according to claim 14, wherein the distance of the vector from the centre of the closest neuron is divided by a square root of the eigenvalue corresponding to the eigenpaxel considered.

16. The method according to claim 15, wherein results corresponding to a same object from different neural networks are added together, and the object indicated as most probable has more than double the distance from a second object that has received a second most points, the object having the most points is recognized.

17. The method as claimed in claim 13, further comprising:

inserting a third object represented by at least one third image;

providing a third radius within said space of N dimensions; and

dimensioning said third radius such that it does not interfere with said first radius.

18. The method as claimed in claim 13, wherein said RBF (radial basis function) neural network comprises only one input layer and only one output layer.

19. The method as claimed in claim 13, further comprising the step of rescaling said first and second filtered matrices, by reducing their size by means of a bilinear interpolation.

20. The method as claimed in claim 13, wherein said objects are faces of individuals.