Automated image interpretation and retrieval system
First Claim
Patent Images
1. A method of classifying a digital image, said method comprising the steps of:
- segmenting the digital image into a plurality of substantially homogeneous regions;
processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacency between the plurality of regions of the digital image;
labeling at least one of the regions of the region adjacency graph with one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph;
providing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image;
analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph;
assigning corresponding ones of the plural stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and
storing the assigned stereotypes and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotype.
1 Assignment
0 Petitions
Accused Products
Abstract
A method of classifying a digital image is disclosed. The method provides a region adjacency graph (310) representing the digital image and analyses the region adjacency graph (310) for predetermined patterns. For each identified pattern a classification of the digital image is selected. The region adjacency graph (310) is classified as one of a number of stereotypes.
78 Citations
88 Claims
-
1. A method of classifying a digital image, said method comprising the steps of:
-
segmenting the digital image into a plurality of substantially homogeneous regions; processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacency between the plurality of regions of the digital image; labeling at least one of the regions of the region adjacency graph with one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph; providing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image; analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph; assigning corresponding ones of the plural stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and storing the assigned stereotypes and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotype. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. An apparatus for classifying a digital image, said apparatus comprising:
-
segmenting means for segmenting the digital image into a plurality of substantially homogeneous regions; processing means for processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacencies between the plurality of regions of the digital image; labeling means for labeling at least one of the regions of the region adjacency graph with one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph; pattern storage means for storing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image; analyzing means for analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph; assigning means for assigning corresponding ones of the plural stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and storage means for storing the assigned stereotypes and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotypes. - View Dependent Claims (17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30)
-
-
31. A computer program product comprising a computer readable medium having a computer program recorded for classifying a digital image, said computer program product comprising:
-
a segmenting module, for segmenting the digital image into a plurality of substantially homogeneous regions; a processing module, for processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacency between the plurality of regions of the digital image; a labeling module, for labeling at least one of the regions of the region adjacency graph with one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph; a pattern storage module for storing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image; an analyzing module, for analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph; an assigning module, for assigning corresponding ones of the plural stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and a storage module, for storing the assigned stereotypes and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotype. - View Dependent Claims (32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45)
-
-
46. A method of classifying a digital image signal, said method comprising the steps of:
-
segmenting the digital image into a plurality of substantially homogeneous regions; processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacency between the plurality of regions of the digital image; labeling at least one of the regions of the region adjacency graph with one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph; providing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image, and wherein each predetermined pattern comprises; (J) a set of labeled regions;
or(ii) a set of labeled regions and corresponding adjacency information; analyzing the labeled region adjacency graph to identify the presence of one or more of the plural patterns of semantic labels in the labeled region adjacency graph; for each pattern of labeled regions identified in the labeled region adjacency graph as matching a predetermined pattern, selecting the corresponding stereotype classifications from the plural stereotype classifications based on the matching; assigning the selected stereotype classifications to the digital image such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and storing the assigned stereotype classifications and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotype classification. - View Dependent Claims (47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58)
-
-
59. An apparatus for classifying a digital image signal, said apparatus comprising:
-
segmenting means for segmenting the digital image into a plurality of substantially homogeneous regions; processing means for processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacency between the plurality of regions of the digital image; labeling means for labeling at least one of the regions of the region adjacency graph with one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph; pattern storage means means for storing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image, and wherein each predetermined pattern comprises; (I) a set of labeled regions;
or(ii) a set of labeled regions and corresponding adjacency information; analyzing means for analyzing the labeled region adjacency graph to identify the presence of one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph, wherein for each pattern of labeled regions identified in the labeled region adjacency graph as matching a predetermined pattern, said classification providing means assigns the predetermined stereotype classifications to the digital image such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and storage means for storing the assigned stereotype classifications and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotype classification. - View Dependent Claims (60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71)
-
-
72. A computer program product comprising a computer readable medium having a computer program recorded for classifying a digital image signal, said computer program product comprising:
-
a segmenting module, for segmenting the digital image into a plurality of substantially homogeneous regions; a processing module, for processing the plurality of regions to provide a region adjacency graph for the digital image, the region adjacency graph representing spatial adjacency between the plurality of regions of the digital image; labeling at least one of the regions of the region adjacency graph with at least one of a plurality of predetermined semantic labels to provide a labeled region adjacency graph; a pattern storage module, for storing a set of plural predetermined patterns of semantic labels, wherein each pattern of semantic labels corresponds to one of plural stereotypes each representing a different classification of a digital image, wherein each predetermined pattern comprises; (I) a set of labeled regions;
or(ii) a set of labeled regions and corresponding adjacency information; an analyzing module, for analyzing the labeled region adjacency graph to identify the presence of one or more of the plural patterns of semantic labels in the labeled region adjacency graph, wherein for each pattern of labeled regions identified in the labeled region adjacency graph as matching a predetermined pattern, said classification providing module assigns the corresponding predetermined stereotype classifications to the digital image such that the assigned stereotype describes the plurality of regions of the digital image and represents a classification of the digital image; and a storage module, for storing the assigned stereotype classifications and the digital image in one or more databases of digital images, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotype classification. - View Dependent Claims (73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84)
-
-
85. A method of classifying a digital image, said method comprising the steps of:
-
segmenting the digital image into substantially homogeneous regions; processing the regions to provide a labeled region adjacency graph for the digital image, wherein the labeled region adjacency graph represents spatial adjacency between the regions of the digital image, and wherein at least one of the regions of the labeled region adjacency graph is associated with one of a plurality of predetermined semantic labels; providing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes based on a minimum size of the regions in the labeled adjacency graph, and wherein the plural stereotypes define a multi-level hierarchical structure with each of the stereotypes representing a different classification of a digital image; analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph; assigning corresponding ones of the plural stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that the assigned stereotype represents a classification of the digital image; and storing the assigned stereotypes and the digital image in one or more databases of digital images together with one or more hierarchical paths, the one or more hierarchical paths being based on the multi-level hierarchical structure, wherein the digital image is retrievable from the one or more databases using a search for the assigned stereotypes and hierarchical paths.
-
-
86. A method of classifying a digital image, said method comprising the steps of:
-
segmenting the image into substantially homogeneous regions; processing, the regions to provide a labeled region adjacency graph for the digital image, wherein the labeled region adjacency graph represents spatial adjacency between the regions of the image, and wherein at least one of the regions of the labeled region adjacency graph is associated with one of a plurality of predetermined semantic labels; providing a set of plural predetermined patterns of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes based on a minimum size of the regions in the labeled adjacency graph, and wherein the plural stereotypes define a multi-level hierarchical structure with each of the stereotypes representing a different classification of a digital image; analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph; and assigning corresponding ones of the plurality of stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that each assigned stereotype represents a classification of the digital image, wherein the plurality of stereotypes are represented in a multi-level hierarchical tree-structure such that the digital image is retrievable from one or more databases of digital images upon selection of at least one of the assigned stereotypes using the multi-level hierarchical tree-structure.
-
-
87. A method of classifying a digital image, said method comprising the steps of:
-
segmenting the image into substantially homogeneous regions; processing the regions to provide a labeled region adjacency graph for the digital image, wherein the labeled region adjacency graph represents spatial adjacency between the regions of the image, and wherein at least one of the regions of the labeled region adjacency graph is associated with one of a plurality of predetermined semantic labels; providing a set of plural predetermined patterns, of semantic labels, wherein each predetermined pattern of semantic labels corresponds to one of plural stereotypes based on a minimum size of the regions in the labeled adjacency graph, and wherein the plural stereotypes define a multi-level hierarchical structure with each of the stereotypes representing a different classification of a digital image; analyzing the labeled region adjacency graph to identify one or more of the plural predetermined patterns of semantic labels in the labeled region adjacency graph; and assigning corresponding ones of the plural stereotypes to the digital image in accordance with which of the plural predetermined patterns were identified in said analyzing step, such that each assigned stereotype represents a classification of the digital image, wherein the plurality of stereotypes have a hierarchical arrangement adapted for representation as a navigable stereotype tree such that the digital image is retrievable from one or more databases of digital images upon selection of at least one of the assigned stereotypes using the stereotype tree.
-
-
88. A method of classifying a digital image, said method comprising the steps of:
-
determining an initial probability value for substantially each of a plurality of predetermined semantic labels, each of said initial probability values being set to a predetermined default value for each of said semantic labels; adjusting said initial probability values for said semantic labels in a selected application domain for said digital image; segmenting said digital image into a plurality of substantially homogeneous regions; processing said plurality of regions to provide a region adjacency graph for the digital image, said region adjacency graph representing spatial adjacency between said plurality of regions of said digital image; labeling each of said regions of said region adjacency graph with a corresponding one of said plurality of predetermined semantic labels to provide a labeled region adjacency graph comprising one or more labeled regions, said one semantic label being selected by multiplying an adjusted initial probability value for each semantic label with a conditional probability density value determined for said at least one region; analyzing said label region adjacency graph to identify one of a plurality of predetermined patterns of labeled regions in said labeled region adjacency graph; assigning one of said plurality of predetermined stereotypes to said labeled region adjacency graph according to said identified predetermined pattern, each of said plurality of predetermined patterns corresponding to one of said predetermined stereotypes such that the assigned stereotype describes the labeled regions of said digital image and represents at least one classification of said digital image; and storing the assigned stereotype and said digital image in one or more databases of digital images, wherein said digital image is retrievable from said one or more databases using the assigned stereotype.
-
Specification