Dynamic partial function in measurement of similarity of objects
A method of measuring similarity of a first object represented by first set of feature values to a second object represented by a second set of feature values, comprising determining respective feature distance values between substantially all corresponding feature values of the first and second sets of feature values, selecting a subset of the determined feature distance values in which substantially all feature distance values that are selected to be within the subset are smaller in value than feature distance values that are not selected to be within the subset, and summing the feature distance values in the subset to produce a partial feature distance measure between the first and second objects.
1. A method of measuring similarity of a first object represented by first set of feature values to a second object represented by a second set of feature values, comprising:
determining respective feature distance values between substantially all corresponding feature values of the first and second sets of feature values;
selecting a subset of the determined feature distance values in which substantially all feature distance values that are selected to be within the subset are smaller in value than feature distance values that are not selected to be within the subset; and
summing the feature distance values in the subset to produce a partial feature distance measure between the first and second objects.
2. The method of claim 1 further including:
scaling the partial feature distance measure to achieve maximum separation between similar and different objects.
3. A method of measuring similarity of a first object X represented by feature values {x 1 , x 2 , x 3 , . . . x p } in a feature set, to a second object Y represented by second feature values in the feature set {y 1 , y 2 , y 3 , . . . y p }, comprising:
determining respective feature distance values δ i between substantially all corresponding feature values x i and y i in accordance with a relationship,
δ i =|x i −y i |
herein x i and y i are respective feature values for the respective first and second objects of the ith feature of the feature set;
selecting a subset Δ m including feature distance values in which substantially all feature distance values that are selected to be within the subset are smaller in value than feature distance values that are not selected to be within the subset; and
producing a partial feature distance measure between the first and second objects by combining the selected feature distance values in the subset Δ m in accordance with a relationship,
d
(
m
,
r
)
=
(
∑
δ
i
εΔ
m
δ
i
r
)
1
r
where r is a scaling factor, and m is the number of feature values in the subset Δ m .
4. The method of claim 3 wherein the first and second objects include respective first and second images.
5. The method of claim 3 wherein, the first and second objects include respective first and second images; and the feature set includes color features and texture features.
6. The method of claim 3 wherein,
the first and second objects include respective first and second images; and the feature set includes color features and texture features and shape features.
7. The method of claim 3 wherein the first and second objects include respective first and second video frames.
8. The method of claim 3 wherein,
the first and second objects include respective first and second video frames; and the feature set includes color features and texture features.
9. The method of claim 3 wherein,
the first and second objects include respective first and second video frames; and
the feature set includes color features and texture features and shape features.
10. The method of claim 3 ,
wherein selecting a subset Δ m includes selecting a prescribed number of features distance values, m, to be within the subset; and
wherein m<p.
11. The method of claim 3 ,
wherein selecting a subset Δ m includes only selecting respective feature distance values that satisfy a prescribed threshold to be within the subset.
12. A method of measuring similarity of a target object X represented by feature values {x 1 , x 2 , x 3 , . . . x p } in a feature set, to a each of multiple respective sample objects {Y 1 , . . . Y N } represented by respective second feature values in the feature set {y 1 , y 2 , y 3 , . . . y p }, comprising: determining feature respective distance values δ i between substantially all corresponding feature values x i and y i in accordance with a relationship,
δ i =|x i −y i |
wherein x 1 and y 1 are respective feature values for the respective first and second objects of the ith feature of the feature set;
identifying respective feature distance value subsets {Δ m1 , . . . Δ mN } corresponding to respective sample objects {Y 1 , . . . Y N };
wherein substantially all feature distance values that are selected to be within a respective feature distance value subset corresponding to a respective sample object are smaller in value than respective feature distance values corresponding to that same respective sample object that are not selected to be within the respective feature distance value subset for that respective sample object; and
producing respective partial feature distance measures between the target object X and respective sample objects {Y 1 , . . . Y N } by combining respective selected feature distance values of the respective feature distance value subsets {Δ m1 , . . . Δ mN } in accordance with a relationship,
d
(
m
,
r
)
=
(
∑
δ
i
εΔ
m
δ
i
r
)
1
r
where r is a scaling factor, and m is the number of feature values in the subset Δ m .
13. The method of claim 12 wherein the target object includes an image and the respective sample objects include images.
14. The method of claim 12 wherein, the target object includes an image and the respective sample objects include images; and the feature set includes color features and texture features.
15. The method of claim 12 wherein, the target object includes an image and the respective sample objects include images; and the feature set includes color features, texture features and shape features.
16. The method of claim 12 wherein the target object includes an image and the respective sample objects include video frames.
17. The method of claim 12 wherein,
the target object includes an image and the respective sample objects include video frames; and the feature set includes color features and texture features.
18. The method of claim 12 wherein,
the target object includes an image and the respective sample objects include video frames; and the feature set includes color features, texture features and shape features.
19. The method of claim 12 ,
wherein selecting a subset Δ m includes selecting a prescribed number of features distance values, m, to be within the subset; and wherein m<p.
20. The method of claim 12 ,
wherein selecting a subset Δ m includes only selecting respective feature distance values that satisfy a prescribed threshold to be within the subset.
21. A process to determine an optimal number of object features to use in comparing a first object and a second object:
a. providing a multiplicity of respective seed objects;
b. providing a respective set of p feature values for each respective seed object;
c. providing a plurality of respective transformations for each respective seed object;
d. providing a respective set of p feature values for each respective transformation of each respective seed object;
e. for each of a multiplicity of seed object selections and for each of a plurality of different values for m,
i. selecting a respective seed object;
ii. selecting a respective value of m<p;
iii. producing respective first partial distance measures between respective feature values of a respective set of p feature values for the respective selected seed object and respective feature values of respective transformations of that same respective selected seed object in accordance with the relationship,
d
(
m
,
r
)
1
=
(
∑
δ
i
εΔ
m1
δ
i1
r
)
1
r
where Δ m1 represents constituents of a set of the m smallest feature distance values δ i1 , for the first partial distance measure, and r is a scaling factor; and
iv. producing respective second partial distance measures between respective feature values of the respective set of p feature values for the respective selected seed object and respective feature values of respective sets of p feature values produced for respective transformations of a multiplicity of the other respective seed objects, in accordance with the relationship,
d
(
m
,
r
)
2
=
(
∑
δ
i
εΔ
m2
δ
i2
r
)
1
2
where Δ m2 represents constituents of a set of the m smallest feature distance values δ i2 , for the first partial distance measure, and r is a scaling factor; and
f. choosing a value of m that on average produces first partial distance measures that are less than second partial distance measures and that on average produces a largest separation between first and second partial distance measures.
22. A process to determine an optimal number of object features to use in comparing a first object and a second object:
a. providing a multiplicity of respective seed objects;
b. providing a respective set of p feature values for each respective seed object;
c. providing a plurality of respective transformations for each respective seed object;
d. providing a respective set of p feature values for each respective transformation of each respective seed object;
e. for each of a multiplicity of seed object selections and for each of a plurality of different threshold feature distance values,
i. selecting a respective seed object;
ii. selecting a respective threshold feature distance value; and
iii. producing respective first partial distance measures between respective feature values of a respective set of p feature values for the respective selected seed object and respective feature values of respective transformations of that same respective selected seed object in accordance with the relationship,
d
(
m
,
r
)
1
=
(
∑
δ
i
ΣΔ
m1
δ
i1
r
)
1
r
where Δ m1 represents constituents of a set including only those feature distance values δ i1 , that satisfy the respective threshold distance feature value, and r is a scaling factor;
iv. producing respective second partial distance measures between respective feature values of the respective set of p feature values for respective seed object and respective feature values of respective sets of p feature values produced for respective transformations of a multiplicity of the other respective seed objects, in accordance with the relationship,
d
(
m
,
r
)
2
=
(
∑
δ
i
εΔ
m2
δ
i2
r
)
1
2
where Δ m2 represents constituents of a set including only those feature distance values δ i2 , that satisfy the respective threshold distance feature value, and r is a scaling factor; and
f. choosing a threshold value that on average produces first partial distance measures that are less than second partial distance measures and that on average produces a largest separation between first and second partial distance measures.