Image retrieval and authentication using enhanced expectation maximization (EEM)
View Patent ↗Technologies are generally presented for employing enhanced expectation maximization (EEM) in image retrieval and authentication. Using uniform distribution as initial condition, the EEM may converge iteratively to a global optimality. If a realization of the uniform distribution is used as the initial condition, the process may also be repeatable. In some examples, a positive perturbation scheme may be used to avoid boundary overflow, often occurring with the conventional EM algorithms. To reduce computation time and resource consumption, a histogram of one dimensional Gaussian Mixture Model (GMM) with two components and wavelet decomposition of an image may be employed.
1. A method for employing an Enhanced Expectation Maximization (EEM) to determine data uniqueness and similarity, the method comprising:
receiving data to be analyzed;
employing, a realization of uniform distribution in an initialization of the EEM on the received data to prevent a convergence to a local maximum and to allow another convergence to a global maximum;
employing a positive perturbation scheme to avoid boundary overflow;
generating a signature vector for the analyzed data and
employing the signature vector to determine one of uniqueness and similarity of the analyzed data to other data based on a similarity of the signature vector to signature vectors of the other data.
2. The method according to claim 1 , wherein the analyzed data is image data.
3. The method according to claim further comprising
employing a histogram of one dimensional Gaussian Mixture Model (GMM) for the image data.
4. The method according to claim 3 , further comprising employing two components of the GM M for histogram.
5. The method according to claim 3 , further comprising:
applying a multilevel discrete wavelet transform to determine as number of the GMMs to be used; and
selecting one or more parameters of the GMM of image wavelet subbands as the signature vector wherein the selected parameters are a mean and a variance of the GMM.
6. The method according to claim 2 , further comprising
determining the uniqueness of the image data based on a Euclidian distance between two signature vectors being greater than a predefined limit.
7. The method according to claim 2 , further comprising
determining the similarity of the image data based on a Euclidian distance between two signature vectors being less than another predefined limit.
8. The method according to claim 2 , wherein a size of the signature vector is 128 bits or less.
9. The method according to claim 2 , further comprising
employing the similarity of the image data for one of image retrieval and indexing; and
employing the uniqueness of the image data for authentication.
10. The method according to claim 2 , further comprising
generating the signature vector by applying the EEM to color components of the image data, wherein a color space of the image data includes one of HSV, YCrCb, CMYK, and RGB.
11. The method according to claim 2 , wherein the positive perturbation scheme includes addition of a denominator factor in goodness calculation of a histogram for the image data and the goodness of the histogram for the image data is determined by:
G=10 log 10 {ee+A/(ee+MSE)}, where
A={Max[Max(h 1 ),Max(h)]} 2 with h and h 1 being histogram values before and after application of the EEM,
MSE=S(h 1 −h) 2 /M with M being a number of occurrence values in the histogram, and
ee is a very small positive perturbation value.
12. The method according to claim 11 , wherein ee is about 10 −20.
13. The method according to claim 1 , further comprising
generating the signature vector by applying the EEM to one of video data and audio data.
14. A computing device for employing an Enhanced Expectation Maximization (EEM) to determine data uniqueness and similarity, the computing device comprising:
a memory configured to store instructions; and
a processor coupled to the memory and configured to execute a data processing application in conjunction with the instructions stored in the memory, wherein the data processing application is configured to:
receive data to be analyzed;
employ a realization of uniform distribution in an initialization of the EEM on the received data to prevent convergence to a local maximum and to allow another convergence to a global maximum;
employ a positive perturbation scheme to avoid boundary overflow;
generate a signature vector fir the analyzed data; and
employ the signature vector to determine one of uniqueness and similarity of the analyzed data to other data based on a similarity of the signature vector to signature vectors of the other data.
15. The computing device according to claim 14 , wherein the analyzed data is image data.
16. The computing device according to claim 15 , wherein the data processing application is further configured to
employ a histogram of one dimensional Gaussian Mixture Model (GMM) for the image data.
17. The computing device according to claim 16 , wherein the data processing application is further configured to
apply a multilevel discrete wavelet transform to determine a number of the GMMs to be used; and
select one or more parameters of the GMM of image wavelet subbands as the signature vector wherein the selected parameters are a meal and a variance of the GMM.
18. The computing device according to claim 14 , wherein the data processing application is further configured to
generate the signature vector by applying the EEM to one of video data and audio data.
19. A non-transitory computer-readable storage medium having instructions stored thereon for employing an Enhanced Expectation Maximization (EEM) to determine data uniqueness and similarity, the instructions comprising;
receiving, data to be analyzed;
employing a realization of uniform distribution in an initialization of the EEM on the received data to prevent a convergence to a local maximum and to allow another convergence to a global maximum;
employing, a positive perturbation scheme to avoid boundary overflow;
generating a signature vector for the analyzed data; and
employing the signature vector to determine one of uniqueness and similarity of the analyzed data to other data based on a similarity of the signature vector to signature vectors of the other data.
20. The non-transitory computer-readable storage medium according to claim 19 , wherein the analyzed data is image data.
21. The non-transitory computer-readable storage medium according to claim 20 , wherein the positive perturbation scheme includes addition of a denominator factor in goodness calculation of a histogram for the image data.
22. The non-transitory computer-readable storage medium according to claim 21 , wherein the goodness of the histogram for the image data is determined by:
G=10 log 10 {ee+A/(ee+MSE)}, where
A={Max[Max(h 1 )Max(h)]} 2 with h and h 1 being histogram values before and after application of the EEM,
MSE=S(h 1 −h) 2 /M with M being a number of occurrence values in the histogram, and
ee is a very small positive perturbation value.
23. The non-transitory computer-readable storage medium according to claim 22 , wherein ee is about 10 −20 .