US6704725B1
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method of searching multimedia data
Assignee: LG Electronics Inc
•
Filing: 2000-06-29
•
Granted: 2004-03-09
•
Expires: 2020-06-29
Abstract
Abstract A method of searching multimedia data is disclosed. In the present invention, an image is searched based upon normalized feature information which has been normalized. By normalizing the feature information, the weights of features used in searching an image can be more efficiently determined, thereby allowing a more accurate image search.
Inventors
- 👤 Jin Soo Lee
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 11:30
CPC Classifications
G; G06; G06F; G06F16/00; G06F16/50; G06F16/58; G06F16/583; G06F16/5838; G06V; G06V10/00; G06V10/70; G06V10/74; G06V10/75; Y; Y10; Y10S; Y10S707/00; Y10S707/99931; Y10S707/99933; Y10S707/99934
Claims Analysis
Total Claims: 32
Claim 1
Independent Claim
115 words
A method of searching multimedia data comprising: (a1) obtaining normalization information of each of a plurality of features in a reference image using at least one multimedia data in a search system, wherein the normalization information has differing values for each of at least two features that have different similarity distributions with respect the reference image; (b1) searching for multimedia data similar to the reference image using at least one of the features and outputting similar multimedia data, wherein weights of features are used; and (c1) normalizing similarity values of the plurality of features of said similar multimedia data using the normalization information, wherein said similarity values are determined based on features used in (b1).
Claim 2
Independent Claim
115 words
A method of searching multimedia data comprising: (a1) obtaining normalization information of each of a plurality of features in a reference image using at least one multimedia data in a search system, wherein the normalization information has differing values for each of at least two features that have different similarity distributions with respect the reference image; (b1) searching for multimedia data similar to the reference image using at least one of the features and outputting similar multimedia data, wherein weights of features are used; and (c1) normalizing similarity values of the plurality of features of said similar multimedia data using the normalization information, wherein said similarity values are determined based on features used in (b1).
Claim 3
Dependent Claim
34 words
A method of claim 1 , further comprising assigning weights of features if weights do not exist or updating existing weights of features, wherein said weights are assigned or updated using the normalized information.
Claim 4
Dependent Claim
36 words
A method of claim 2 , further comprising receiving user feedback after searching for the reference image in (b1), and assigning or updating the weights of features using both the normalization information and the user feedback.
Claim 5
Dependent Claim
21 words
A method of claim 1 , further comprising sorting the normalized multimedia data based upon relevance and displaying sorted multimedia data.
Claim 6
Dependent Claim
57 words
A method of claim 1 , wherein (a1) comprises: (a2) obtaining a probability density function (PDF) of each feature based upon a similarity between the reference image and said at least one multimedia data; (b2) obtaining a cumulative distribution function (CDF) of each PDF; and (c2) obtaining normalizing information for the reference image based upon each CDF.
Claim 7
Dependent Claim
42 words
A method of claim 5 , wherein (c2) comprises: obtaining a slope s and a mean m to approximate each CDF using a Sigmoid function; and setting a slope s′ and a mean m′ of a reference CDF representing the reference image.
Claim 8
Dependent Claim
36 words
A method of claim 6 , wherein obtaining the slope s and the mean m by calculating an error between the actual CDF and an approximated CDF using different values of slope s and mean m.
Claim 9
Dependent Claim
56 words
A method of claim 6 , wherein setting the slope s′ and the mean m′ based upon a mean value of the slopes and a mean value of the means of each feature, a slope and mean values of a CDF for a normal distribution, or a slope and a mean value set by a user.
Claim 10
Dependent Claim
33 words
A method of claim 6 , wherein in (c1), normalizing a similarity value based upon a feature using a corresponding slope s, a corresponding mean m, the slope s′ and the mean m′.
Claim 11
Dependent Claim
19 words
A method of claim 1 , wherein the normalization information is different for the features in the reference image.
Claim 12
Independent Claim
85 words
A method of normalizing similarity values between a reference image and target images based on features, comprising: (a1) obtaining normalization information of features in a reference image using at least one target image in a search system; and (b1) obtaining similar values based on similar target images; (c1) normalizing the similarity values resulting from a search using the normalization information, wherein the normalization information has differing values for each of at least two features that have different similarity representations with respect to the reference image.
Claim 13
Independent Claim
85 words
A method of normalizing similarity values between a reference image and target images based on features, comprising: (a1) obtaining normalization information of features in a reference image using at least one target image in a search system; and (b1) obtaining similar values based on similar target images; (c1) normalizing the similarity values resulting from a search using the normalization information, wherein the normalization information has differing values for each of at least two features that have different similarity representations with respect to the reference image.
Claim 14
Dependent Claim
53 words
A method of claim 11 , wherein (a1) comprises: (a2) obtaining a probability density function (PDF) of each feature based upon a similarity between the reference image and said at least one target image; (b2) obtaining a cumulative distribution function (CDF) of each PDF; and (c2) obtaining normalizing information based upon each CDF.
Claim 15
Dependent Claim
42 words
A method of claim 12 , wherein (c2) comprises: obtaining a slope s and a mean m to approximate each CDF using a Sigmoid function; and setting a slope s′ and a mean m′ of a reference CDF of the reference image.
Claim 16
Dependent Claim
36 words
A method of claim 13 , wherein obtaining the slope s′ and the mean m by calculating an error between the actual CDF and an approximated CDF using different values of slope s and mean m.
Claim 17
Dependent Claim
56 words
A method of claim 13 , wherein setting the slope s′ and the mean m′ based upon a mean value of the slopes and a mean value of the means of each feature, a slope and mean values of a CDF for a normal distribution, or a slope and a mean value set by a user.
Claim 18
Dependent Claim
33 words
A method of claim 13 , wherein in (c1), normalizing a similarity value based upon a feature using a corresponding slope s, a corresponding mean m, the slope s′ and the mean m′.
Claim 19
Dependent Claim
39 words
A method of claim 11 , wherein the normalization information is different for the features in the reference image, and wherein each feature is evaluated differently when determining the similarity values, and wherein the target image comprises multimedia data.
Claim 20
Dependent Claim
60 words
A data structure stored in a computer readable medium for an image in an image search system comprising: features descriptor of a reference image; and normalization information corresponding to each feature in the feature descriptor, wherein the normalization information has differing values for each of at least two features that have different similarity distributions with respect to the reference image.
Claim 21
Dependent Claim
60 words
A data structure stored in a computer readable medium for an image in an image search system comprising: features descriptor of a reference image; and normalization information corresponding to each feature in the feature descriptor, wherein the normalization information has differing values for each of at least two features that have different similarity distributions with respect to the reference image.
Claim 22
Dependent Claim
25 words
A data structure stored in a computer readable medium of claim 18 , wherein the normalization information comprises a slope information and a mean information.
Claim 23
Dependent Claim
70 words
A data structure stored in a computer readable medium of claim 18 , wherein the normalization information is obtained by a method comprising: (a) obtain a probability density function (PDF) of each feature in the feature descriptor based upon a similarity between a reference image and at least one target image; (b) obtaining a cumulative distribution function (CDF) of each PDF; and (c) obtaining normalizing information based upon each CDF.
Claim 24
Dependent Claim
45 words
A data structure stored in a computer readable medium of claim 20 , wherein (c) comprises: obtaining a slope s and a mean m to approximate each CDF using a Sigmoid function; and setting a slope s′ and a mean m′ of a reference CDF.
Claim 25
Dependent Claim
26 words
A data structure stored in a computer readable medium of claim 18 , wherein the normalization information is different for the features in the reference image.
Claim 26
Independent Claim
204 words
A method of searching multimedia data comprising: selecting a reference image to search; preparing an initial set of a plurality of target images to obtain a similarity distribution of each of a plurality of features; determining a similarity probability density function (PDF) of each of said plurality of features in the reference image by comparing the similarity of said each of the plurality of features in the reference image and the plurality of target images; determining a cumulative distribution function (CDF) of said each of the plurality of features using a corresponding similarity PDF; determining a normalized CDF for said each of the plurality of features, wherein at least two selected features of the plurality of features have different normalized CDFs based on said similarity distributions; conducting a search of multimedia data using selected features of the reference image, wherein the selected features have corresponding weights; receiving relevance feedback for images in an image list returned by the search; determining a similarity between each feedback image and the reference image using the at least two selected features; normalizing the similarity for said at least two selected features using the corresponding normalized CDFs; and updating the weights of the selected features using the normalized similarities.
Claim 27
Independent Claim
204 words
A method of searching multimedia data comprising: selecting a reference image to search; preparing an initial set of a plurality of target images to obtain a similarity distribution of each of a plurality of features; determining a similarity probability density function (PDF) of each of said plurality of features in the reference image by comparing the similarity of said each of the plurality of features in the reference image and the plurality of target images; determining a cumulative distribution function (CDF) of said each of the plurality of features using a corresponding similarity PDF; determining a normalized CDF for said each of the plurality of features, wherein at least two selected features of the plurality of features have different normalized CDFs based on said similarity distributions; conducting a search of multimedia data using selected features of the reference image, wherein the selected features have corresponding weights; receiving relevance feedback for images in an image list returned by the search; determining a similarity between each feedback image and the reference image using the at least two selected features; normalizing the similarity for said at least two selected features using the corresponding normalized CDFs; and updating the weights of the selected features using the normalized similarities.
Claim 28
Independent Claim
107 words
A method of searching multimedia data comprising: extracting feature information and normalizing information for a plurality of features in a reference image; determining a set of target images; setting a first image of the target images as an object image; determining a similarity of each of a plurality of features in the object image and the reference image; and normalizing the similarity of said each of the plurality of features using corresponding normalization information, wherein at least two of the features have different similarity distributions with respect to the reference image, and wherein the normalization information is different for said each of the at least two features.
Claim 29
Independent Claim
107 words
A method of searching multimedia data comprising: extracting feature information and normalizing information for a plurality of features in a reference image; determining a set of target images; setting a first image of the target images as an object image; determining a similarity of each of a plurality of features in the object image and the reference image; and normalizing the similarity of said each of the plurality of features using corresponding normalization information, wherein at least two of the features have different similarity distributions with respect to the reference image, and wherein the normalization information is different for said each of the at least two features.
Claim 30
Dependent Claim
29 words
The method of claim 24 , comprising updating weights of said each of the plurality of features using the normalized similarity of said each of the plurality of features.
Claim 31
Dependent Claim
47 words
The method of claim 25 , comprising: setting a next image of the set of target images as the object image and repeating the normalizing the similarity of said each of the plurality of features until a last image of the set of target images is selected.
Claim 32
Dependent Claim
45 words
The method of claim 26 , comprising: determining a whole image similarity by multiplying feature weights by the normalized similarity for said each of the plurality of features; adding the multiplied values; and ranking the set of target images by each corresponding whole image similarity.
References Cited (Backward Citations)
US5742706A
Cited
US5802361A
Cited
US5855015A
Cited
US5893095A
Cited
US6424746B1
Cited
US6269358B1
Cited
US6567551B2
Cited
US6263334B1
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2007-03-09 | 2007-09-08 | 2008-03-09 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2011-03-09 | 2011-09-08 | 2012-03-09 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2015-03-09 | 2015-09-08 | 2016-03-09 | $7700.00 | Window Closed / Lapsed |