Patent US5956427A successfully refreshed from Google Patents.
US5956427A
Expired / Lapsed
US
⏱️ 0.0 years remaining
DFT encoding of oriented filter responses for rotation invariance and orientation estimation in digitized images
Assignee: California Institute of Technology
•
Filing: 1995-06-15
•
Granted: 1999-09-21
•
Expires: 2015-06-15
Abstract
Abstract A technique of determining information about orientation and amount of orientation of an image. The orientations of the image are sampled, and the samples are used to reconstruct information from which all orientations can be obtained. The samples each include information about specific components of the image, and these samples are assembled into a feature vector. The feature vector is discrete Fourier transformed whereby its continuous orientation information can be obtained.
Inventors
- 👤 Hayit Greenspan
- 👤 Serge Belongie
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 12:14
CPC Classifications
G; G06; G06T; G06T7/00; G06T7/70; G06V; G06V10/00; G06V10/40; G06V10/44; G06V10/443
Claims Analysis
Total Claims: 25
Claim 1
Independent Claim
173 words
A device for determining orientation information throughout an image, comprising: a first element operating to produce information indicative of an image to be analyzed; a plurality of oriented filters, each receiving said information, each oriented filter having an orientation setting, and each oriented filter outputting information indicative of a component of said information which has an orientation passband centered at said orientation setting, to produce a set of filtered values, each filtered value including different information than each other filtered value; a memory element, storing said set of filtered values; and a processing device, operating on said set of filtered values in said memory element to produce both rotation invariant information indicating a magnitude of orientation information of said image, and rotation variant information indicating specific local orientations in said image, wherein said processing device operates on said set of filtered values by performing a discrete Fourier transform thereon, to obtain a complex value with a magnitude which indicates said rotation invariant information and with a phase which indicates said rotation variant information.
Claim 2
Independent Claim
173 words
A device for determining orientation information throughout an image, comprising: a first element operating to produce information indicative of an image to be analyzed; a plurality of oriented filters, each receiving said information, each oriented filter having an orientation setting, and each oriented filter outputting information indicative of a component of said information which has an orientation passband centered at said orientation setting, to produce a set of filtered values, each filtered value including different information than each other filtered value; a memory element, storing said set of filtered values; and a processing device, operating on said set of filtered values in said memory element to produce both rotation invariant information indicating a magnitude of orientation information of said image, and rotation variant information indicating specific local orientations in said image, wherein said processing device operates on said set of filtered values by performing a discrete Fourier transform thereon, to obtain a complex value with a magnitude which indicates said rotation invariant information and with a phase which indicates said rotation variant information.
Claim 3
Dependent Claim
14 words
A device as in claim 1 wherein said oriented filters are steerable pyramid filters.
Claim 4
Independent Claim
185 words
A device for determining orientation information throughout an image, comprising: a first element operating to produce information indicative of an image to be analyzed; a plurality of oriented filters, each receiving said information, each oriented filter having an orientation setting, and each oriented filter outputting information indicative of a component of said information which has an orientation passband centered at said orientation setting, to produce a set of filtered values, each filtered value including different information than each other filtered value; a memory element, storing said set of filtered values; and a processing device, operating on said set of filtered values in said memory element to produce both rotation invariant information indicating a magnitude of orientation information of said image, and rotation variant information indicating specific local orientations in said image, wherein said processing device converts said set of filtered values into a frequency domain representation, wherein said processing device forms feature vectors from said filtered values, each said feature vector including a plurality of corresponding points from a plurality of said filtered values, at least one of said feature vectors representing rotation invariant information.
Claim 5
Independent Claim
185 words
A device for determining orientation information throughout an image, comprising: a first element operating to produce information indicative of an image to be analyzed; a plurality of oriented filters, each receiving said information, each oriented filter having an orientation setting, and each oriented filter outputting information indicative of a component of said information which has an orientation passband centered at said orientation setting, to produce a set of filtered values, each filtered value including different information than each other filtered value; a memory element, storing said set of filtered values; and a processing device, operating on said set of filtered values in said memory element to produce both rotation invariant information indicating a magnitude of orientation information of said image, and rotation variant information indicating specific local orientations in said image, wherein said processing device converts said set of filtered values into a frequency domain representation, wherein said processing device forms feature vectors from said filtered values, each said feature vector including a plurality of corresponding points from a plurality of said filtered values, at least one of said feature vectors representing rotation invariant information.
Claim 6
Dependent Claim
37 words
A device as in claim 1 wherein said processing device operates on said set of filtered values to produce a continuous feature vector that includes data which represents a prediction of the filter responses at any orientation.
Claim 7
Dependent Claim
46 words
A device as in claim 4 further comprising a database memory storing information about a plurality of textures, and wherein said processing device compares processed filtered values in the memory element to a plurality of filtered elements in the database to determine a closest match therebetween.
Claim 8
Dependent Claim
36 words
A system as in claim 2 wherein said steerable filters produce multiple output, each output indicative of a particular component resolution of said image, said filtered values including multiple orientations and multiple resolutions of the image.
Claim 9
Dependent Claim
26 words
A system as in claim 2 wherein said filters have orientations each of 0°, 45°, 90°, 135°, and 0.5 octave or 1 octave separation in scale.
Claim 10
Dependent Claim
24 words
A device as in claim 6 wherein said processing device operates on said set of filtered values by performing a discrete Fourier transform thereon.
Claim 11
Independent Claim
185 words
A device for determining orientation information about an image, comprising: a first element operating to produce information indicative of an image to be analyzed; a plurality of oriented filters, each receiving said information, each oriented filter having an orientation setting, and each oriented filter outputting information indicative of a part of said information which has an orientation passband centered at said orientation setting to produce a set of filtered values, each filtered value including different information than each other filtered value; a memory element, storing said set of filtered values; and a processing device, operating on said set of filtered values in said memory element to produce sample information, each item of said sample information including information from said filtered values, and said sample information including information from which other orientations, other than an orientation from which said sample information was obtained, can be determined, wherein said processing device performs a discrete Fourier transform on said set of filtered values by converting the samples along the orientation axis into a frequency domain representation, and wherein said frequency domain representation includes information which is rotation invariant.
Claim 12
Independent Claim
185 words
A device for determining orientation information about an image, comprising: a first element operating to produce information indicative of an image to be analyzed; a plurality of oriented filters, each receiving said information, each oriented filter having an orientation setting, and each oriented filter outputting information indicative of a part of said information which has an orientation passband centered at said orientation setting to produce a set of filtered values, each filtered value including different information than each other filtered value; a memory element, storing said set of filtered values; and a processing device, operating on said set of filtered values in said memory element to produce sample information, each item of said sample information including information from said filtered values, and said sample information including information from which other orientations, other than an orientation from which said sample information was obtained, can be determined, wherein said processing device performs a discrete Fourier transform on said set of filtered values by converting the samples along the orientation axis into a frequency domain representation, and wherein said frequency domain representation includes information which is rotation invariant.
Claim 13
Dependent Claim
53 words
A device as in claim 9 wherein said oriented filters are pyramid filters, each of which is oriented to have a passband centered at a specific orientation setting, and each of which produces a plurality of successively filtered outputs, said sample information including information from each of said plurality of successively filtered outputs.
Claim 14
Dependent Claim
31 words
A device as in claim 10 wherein said sample information includes vectors, each vector indicating a specific area of the image, and each area formed from a plurality of said components.
Claim 15
Dependent Claim
16 words
A device as in claim 11 wherein each vector element also includes information from multiple orientations.
Claim 16
Independent Claim
179 words
A method of determining orientation information about images containing multiple orientations at a point, comprising: producing a signal indicative of an image to be analyzed, said producing step includes acquiring an image with a camera; filtering said signal at orientations centered at a plurality of different settings to obtain a set of filtered values, each filtered value including information about different information than each other filtered value; storing said set of filtered values in memory; operating on said set of filtered values in said memory to produce vector information, each vector information including information from said filtered values; and interpreting said vector information to obtain a continuous signal indicative of a continuous feature waveform indicative of all orientation within said image, wherein said filtering step filters said signal to obtain a plurality of different image components; each said image component being part of said vector, wherein said interpreting comprises changing said vector from the time domain to the frequency domain to obtain frequency invariant information thereof, and wherein said changing comprises discretely Fourier transforming samples of oriented filtered outputs.
Claim 17
Independent Claim
179 words
A method of determining orientation information about images containing multiple orientations at a point, comprising: producing a signal indicative of an image to be analyzed, said producing step includes acquiring an image with a camera; filtering said signal at orientations centered at a plurality of different settings to obtain a set of filtered values, each filtered value including information about different information than each other filtered value; storing said set of filtered values in memory; operating on said set of filtered values in said memory to produce vector information, each vector information including information from said filtered values; and interpreting said vector information to obtain a continuous signal indicative of a continuous feature waveform indicative of all orientation within said image, wherein said filtering step filters said signal to obtain a plurality of different image components; each said image component being part of said vector, wherein said interpreting comprises changing said vector from the time domain to the frequency domain to obtain frequency invariant information thereof, and wherein said changing comprises discretely Fourier transforming samples of oriented filtered outputs.
Claim 18
Dependent Claim
163 words
An element for recognizing a texture, comprising: an image receiving element, receiving information indicative of an image of the texture to be processed; a set of oriented pyramid filters, respectively producing multiple outputs respectively indicative of components of said information at different levels of resolution, to produce filtered information; each part of said information indicative of a particular orientation of an image, and indicative of a resolution of said image lower than an input resolution of said image; and a processing device, processing said filtered information to obtain first and second information therefrom; said first information indicative of an overall magnitude of the information and said second information indicative of the orientation of said information, wherein said processing device carries out a discrete Fourier transform on the information, and obtains and interprets outputs of said discrete Fourier transform as said parts; and said processing device compares said first and second information to other determined values, to recognize a texture in any rotated orientation.
Claim 19
Dependent Claim
163 words
An element for recognizing a texture, comprising: an image receiving element, receiving information indicative of an image of the texture to be processed; a set of oriented pyramid filters, respectively producing multiple outputs respectively indicative of components of said information at different levels of resolution, to produce filtered information; each part of said information indicative of a particular orientation of an image, and indicative of a resolution of said image lower than an input resolution of said image; and a processing device, processing said filtered information to obtain first and second information therefrom; said first information indicative of an overall magnitude of the information and said second information indicative of the orientation of said information, wherein said processing device carries out a discrete Fourier transform on the information, and obtains and interprets outputs of said discrete Fourier transform as said parts; and said processing device compares said first and second information to other determined values, to recognize a texture in any rotated orientation.
Claim 20
Dependent Claim
49 words
A device as in claim 14 wherein said filters are oriented pyramid filters which produce a plurality of lower resolution images and wherein there are at least three of said pyramid filters, each of pyramid filters producing a lower resolution image than a previous one of said pyramid filters.
Claim 21
Dependent Claim
34 words
A device as in claim 15 wherein said processing device produces feature vectors which include information from each of said pyramid filters said feature vectors including information indicating rotation variant and rotation invariant parts.
Claim 22
Independent Claim
165 words
A method of processing an image comprising the steps of: determining components of the image, including at least a first component with coarse details and a second component with finer details at discrete orientations centered at discrete orientation center passbands; storing said first and second components in a memory; arranging contents of said memory into a vector form, said vector including elements from said different resolution components, and elements from said discrete orientation center passbands; and processing said vector to obtain information indicative of orientation center passbands of the image other than said discrete orientation center passband, wherein said processing includes carrying out a conversion on said vector to convert information in said vector from time domain to frequency domain, and to obtain orientation independent information therefrom, and wherein said vector is a feature vector having components whose positions shift along an orientation axis depending on a rotation of said input image; and said time domain to frequency domain conversion is a discrete Fourier transform.
Claim 23
Independent Claim
165 words
A method of processing an image comprising the steps of: determining components of the image, including at least a first component with coarse details and a second component with finer details at discrete orientations centered at discrete orientation center passbands; storing said first and second components in a memory; arranging contents of said memory into a vector form, said vector including elements from said different resolution components, and elements from said discrete orientation center passbands; and processing said vector to obtain information indicative of orientation center passbands of the image other than said discrete orientation center passband, wherein said processing includes carrying out a conversion on said vector to convert information in said vector from time domain to frequency domain, and to obtain orientation independent information therefrom, and wherein said vector is a feature vector having components whose positions shift along an orientation axis depending on a rotation of said input image; and said time domain to frequency domain conversion is a discrete Fourier transform.
Claim 24
Dependent Claim
33 words
A method as in claim 17 wherein one element of the feature vector is a non-oriented element, and other elements of the feature vector are oriented elements which shift depending on the orientation.
Claim 25
Dependent Claim
18 words
A method as in claim 18 wherein said analyzing analyzes harmonics of the feature curve and phase thereof.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2002-09-20 | 2003-03-22 | 2003-09-21 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2006-09-20 | 2007-03-22 | 2007-09-21 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2010-09-20 | 2011-03-22 | 2011-09-21 | $7700.00 | Window Closed / Lapsed |