Method and apparatus for obtaining and processing image features
View Patent ↗Machine-readable media, methods, apparatus and system for obtaining and processing image features are described. In some embodiments, groups of training features derived from regions of training images may be trained to obtain a plurality of classifiers, each classifier corresponding to each group of training features. The plurality of classifiers may be used to classify groups of validation features derived from regions of validation images to obtain a plurality of weights, wherein each weight corresponds to each region of the validation images and indicates how important the each region of the validation images is. Then, a weight may be discarded from the plurality of weights based upon a certain criterion.
1. A method, comprising:
training groups of training features derived from regions of training images to obtain a plurality of classifiers, wherein each classifier corresponds to each group of training features,
utilizing the plurality of classifiers to classify groups of validation features derived from regions of validation images to obtain a plurality of weights, wherein each weight corresponds to each region of the validation images and indicates how important the each region of the validation images is; and
discarding a weight from the plurality of weights based upon a certain criterion, wherein the obtaining the plurality of weights further comprises:
establishing a discriminant function
η
(
z
)
=
log
P
(
t
=
1
❘
z
)
P
(
t
=
0
❘
z
)
=
log
P
(
t
=
1
❘
z
)
1
-
P
(
t
=
1
❘
z
)
=
β
T
z
,
wherein P represents a probability, t=1 represents a validation image of the validation images positively matches to a training image of the training images, t=0 represents the validation image negatively matches to the training image, z represents how close the each region of the validation image is to the each region of the training image, β T represents the plurality of weights;
establishing a likelihood function
L
(
z
)
=
∑
t
(
z
i
)
=
1
η
(
z
i
)
-
∑
t
(
z
j
)
=
0
η
(
z
j
)
,
wherein
∑
t
(
z
i
)
=
1
η
(
z
i
)
represents a sum of discriminant functions for one or more of the validation images that positively match to one or more of the training images, and
∑
t
(
z
j
)
=
0
η
(
z
j
)
represents a sum of discriminant functions for one or more of the validation images that negatively match to one or more of the training images; and
obtaining the plurality of weights by maximizing the likelihood function.
2. The method of claim 1 , wherein the weight is a negative weight.
3. The method of claim 1 , further comprising:
obtaining the groups of training features from the regions of the training images, wherein the each group of training features corresponds to each region of the training images; and
obtaining the groups of validation features from the regions of the validation images, wherein each group of validation features corresponds to the each region of the validation images and the each region of the validation images corresponds to the each region of the training images.
4. The method of claim 1 , wherein the utilizing the plurality of classifiers further comprises:
utilizing the plurality of classifiers to classify the groups of validation features to obtain a plurality of similarity factors, wherein each similarity factor corresponds to the each region of the validation images and indicates how close the each region of the validation images is to a corresponding region of the training images; and
obtaining the plurality of weights based upon the plurality of similarity factors.
5. The method of claim 1 , further comprising:
utilizing a remaining weight from the plurality of weights to optimize the plurality of classifiers, to provide an optimized classifier;
determining a region from a test image with reference to the remaining weight;
obtaining a group of test features from the region of the test image; and
utilizing the optimized classifier to classify the group of test features.
6. A system, comprising:
a processor;
a machine-readable medium coupled to the processor, wherein the machine-readable medium further comprises a plurality of instructions which, if executed by the processor, cause the system to perform operations comprising:
training groups of training features derived from regions of training images to obtain a plurality of classifiers, wherein each classifier corresponds to each group of training features,
utilizing the plurality of classifiers to classify groups of validation features derived from regions of validation images to obtain a plurality of weights, wherein each weight corresponds to each region of the validation images and indicates how important the each region of the validation images is; and
discarding a weight from the plurality of weights based upon a certain criterion, wherein the plurality of instructions that cause the system to obtain the plurality of weights, further causes the system to perform operations comprising:
establishing a discriminant function
η
(
z
)
=
log
P
(
t
=
1
❘
z
)
P
(
t
=
0
❘
z
)
=
log
P
(
t
=
1
❘
z
)
1
-
P
(
t
=
1
❘
z
)
=
β
T
z
,
wherein P represents a probability, t=1 represents a validation image of the validation images positively matches to a training image of the training images, t=0 represents the validation image negatively matches to the training image, z represents how close the each region of the validation image is to the each region of the training image, β T represents the plurality of weights;
establishing a likelihood function
L
(
z
)
=
∑
t
(
z
i
)
=
1
η
(
z
i
)
-
∑
t
(
z
j
)
=
0
η
(
z
j
)
,
wherein
∑
t
(
z
i
)
=
1
η
(
z
i
)
represents a sum of discriminant functions for one or more of the validation images that positively match to one or more of the training images, and
∑
t
(
z
j
)
=
0
η
(
z
j
)
represents a sum of discriminant functions for one or more of the validation images that negatively match to one or more of the training images; and
obtaining the plurality of weights by maximizing the likelihood function.
7. The system of claim 6 , wherein the weight is a negative weight.
8. The system of claim 6 , wherein the plurality of instructions further cause the system to perform operations comprising:
obtaining the groups of training features from the regions of the training images, wherein the each group of training features corresponds to each region of the training images; and
obtaining the groups of validation features from the regions of the validation images, wherein each group of validation features corresponds to the each region of the validation images and the each region of the validation images corresponds to the each region of the training images.
9. The system of claim 6 , wherein the plurality of instructions that cause the system to utilize the plurality of classifiers, further causes the system to perform operations comprising:
utilizing the plurality of classifiers to classify the groups of validation features to obtain a plurality of similarity factors, wherein each similarity factor corresponds to the each region of the validation images and indicates how close the each region of the validation images is to a corresponding region of the training images; and
obtaining the plurality of weights based upon the plurality of similarity factors.
10. The system of claim 6 , the plurality of instructions further cause the system to perform operations comprising:
utilizing a remaining weight from the plurality of weights to optimize the plurality of classifiers, to provide an optimized classifier;
determining a region from a test image with reference to the remaining weight;
obtaining a group of test features from the region of a test image; and
utilizing the optimized classifier to classify the group of test features.