Method for image retrieval with multiple regions of interest
AssigneeMitsubishi Electric Research Laboratories, Inc.
InventorsBaback Moghaddam, Henning Biermann, Dimitris Margaritis
Life of this patent
Estimating the term…
Abstract
A method for representing an image in an image retrieval database first separates and filters images to extract color and texture features. The color and texture features of each image are partitioned into a plurality of blocks. A joint distribution of the color features and a joint distribution of the texture features are estimated for each block. The estimated joint distributions are stored in the database with each image to enable retrieval of the images by comparing the estimated joint distributions.
Claims17
A method for representing an image in an image retrieval database, comprising the steps of:
- filtering the image to extract texture features, wherein the texture features include three edge measurements and wherein the three edge measurements are an edge magnitude, a rotation invariant Laplacian, and an edge orientation;
- separating the image to extract color features, wherein the color features include three color coordinates;
- partitioning the color and texture features into a plurality of blocks;
- estimating a joint distribution of the color features and a joint distribution of the texture features of each block;
- and
- storing the estimated joint distributions in the database with the image;
- filtering and separating a query image to extract color and texture features of the query image;
- partitioning the color and texture features of the query image into a plurality of blocks;
- estimating a joint distribution of the color features and a joint distribution of the texture features of the query image;
- specifying a region of interest in the query image by outlining with a user interface;
- summing the joint distributions of the color and texture features corresponding to the specified region of interest;
- and
- indexing the database with the summed joint distributions to locate a matching image for the query image.
An image retrieval system comprising:
- means for separating and filtering an image to extract color and texture features associated with each image, wherein the texture features include three edge measurements and wherein the three edge measurements are an edge magnitude, a rotation invariant Laplacian, and an edge orientation and wherein the color features include three color coordinates;
- means for partitioning the color and texture features of each image into a plurality of blocks;
- means for estimating a joint distribution of the color features and a joint distribution of the texture features of each block of each image;
- and
- a memory storing the estimated joint distributions with each image;
- means for separating and filtering a query image to extract color and texture features of the query image, and a joint distribution of the color features and a joint distribution of the texture features of the query image is specified;
- a user interface specifying a region of interest in the query image;
- an adder summing the joint distributions of the color and texture features corresponding to the specified region of interest;
- and
- means for indexing the database with the summed joint distributions to locate a matching image for the query image.
An image retrieval system comprising:
- a user interface configured to specify a region of interest in a query image by outlining;
- means for estimating a joint distribution of color features and a joint distribution of texture features of the region of interest, wherein the texture features include three edge measurements and wherein the three edge measurements are an edge magnitude, a rotation invariant Laplacian, and an edge orientation and wherein the color features include three color coordinates;
- an adder summing the joint distributions of the color and texture features corresponding to the region of interest;
- and
- means for indexing a database of stored images with the summed joint distributions to locate a matching image in the database for the query image.
Dependencies are inferred from the claim text.