Multi-scale representation of an out of focus image

1. A method for generating a multi-scale representation of a blurred input image, the blurred input image corresponding to an initial blur Gaussian kernel σ blur , the method comprising the steps of:

estimating, by a scale space level estimator, said initial blur Gaussian kernel σ blur corresponding to said blurred input image;

determining, by a multi-scale representation generator, a set of Gaussian difference kernels according to said initial blur Gaussian kernel σ blur , and according to a predetermined set of Gaussian kernels, each of said set of Gaussian difference kernels being determined according to formula σ diff =√{square root over (σ required 2 −σ blur 2 )} , wherein σ diff is a value of variance of a specific Gaussian difference kernel, and σ required is a value of variance of a selected one of said set of predetermined Gaussian kernels, corresponding to said specific Gaussian difference kernel; and

generating, by the multi-scale representation generator, a multi-scale representation of said input image by applying each of said set of Gaussian difference kernels on said input image.

2. The method according to claim 1 , further comprising the procedure of receiving said input image before said procedure of estimating.

3. The method according to claim 1 , wherein said method further comprising a procedure of performing at least one of feature detection and feature classification, on said multi-scale representation of said input image, after said procedure of generating.

4. The method according to claim 1 , wherein said method further comprising a procedure of performing at least one of object recognition and object classification, on said multi-scale representationof said input image, after said procedure of generating.

5. The method according to claim 1 , wherein said method further comprising a procedure of performing at least one of image classification and shape analysis, on said multi-scale representation of said input image, after said procedure of generating.

6. The method according to claim 1 , wherein said procedure of determining a set of Gaussian difference kernels includes a sub procedure of determining said predeteremined set of Gaussian kernels, and wherein each of the Gaussian kernels of said predetermined set of Gaussian kernels, has scale factor higher than that of a previous Gaussian kernel of said predetermined set of Gaussian kernels.

7. The method according to claim 1 , wherein said procedure of generating said multi-scale representation is performed by applying each Gaussian difference kernel of said set of Gaussian difference kernels to said input image.

8. The method according to claim 1 , wherein said input image is a blurred version of a hypothetic focused image, said input image being substantially similar to an image received by applying a Gaussian kernel, having said estimated scale factor, on said hypothetic focused image.

9. A system for generating a multi-scale representation of a blurred input image, the blurred input image corresponding to an initial blur Gaussian kernel σ blur , the system comprising:

a scale space level estimator for estimating said initial blur Gaussian kernel σ blur corresponding to said blurred input image;

a multi-scale representation generator coupled with said scale space level estimator, for receiving said estimated scale factor, determining a set of Gaussian difference kernels according to said initial blur Gaussian kernel σ blur and according to a predetermined set of Gaussian kernels, and for generating a multi-scale representation of said input image by applying each of said set of Gaussian difference kernels to said input image, wherein said multi-scale representation generator determining of said set of Gaussian difference kernels according to the following formula: σ diff =√{square root over (σ required 2 −σ blur 2 )} ,wherein σ diff is the value of variance of a specific Gaussisn difference kernel, and σ required is the value of variance of a selected one of said set of predetermined Gaussian kernels.

10. The system according to claim 9 , wherein said system further comprises an image source coupled with said scale space level estimator, for providing said input image to said scale space level estimator.

11. The system according to claim 10 , wherein said image source is selected from the list consisting of:

an image capture device;

a storage unit storing the input image; and

a communication interface receiving the input image from an external source.

12. The system according to claim 9 , wherein said system further comprises a visual processor coupled with said multi-scale representation generator, said visual processor performing at least one of feature detection and feature classification, on said multi-scale representation of said input image.

13. The system according to claim 9 , wherein said system further comprises a visual processor coupled with said multi-scale representation generator, said visual processor performing at least one of object recognition and object classification, on said multi-scale representation of said input image.

14. The system according to claim 9 , wherein said system further comprises a visual processor coupled with said multi-scale representation generator, said visual processor performing at least one of image classification and shape analysis, on said multi-scale representation of said input image.

15. The system according to claim 9 , wherein said multi-scale representation generator determines said predetermined set of Gaussian kernels, and wherein each of said predetermined set of Gaussian kernels has scale factor higher than that of a previous Gaussian kernel of said predetermined set of Gaussian kernels.

16. The system according to claim 9 , wherein said multi-scale representation generator generates said multi-scale representation by applying each Gaussian difference kernel of said set of Gaussian difference kernels to said input image.

17. The system according to claim 9 , wherein said input image is a blurred version of a hypothetic focused image, said input image being substantially similar to an image received by applying a Gaussian kernel, having said estimated scale factor, on said hypothetic focused image.