Methods of imaging based on wavelet retrieval of scenes
AssigneeLucent Technologies Inc.
InventorsApostol Ivanov Natsev, Rajeev Rastogi, Kyuseok Shim
Life of this patent
Estimating the term…
Abstract
Methods of imaging objects based on wavelet retrieval of scenes utilize wavelet transformation of plural defined regions of a query image. By increasing the granularity of the query image to greater than one region, accurate feature vectors are obtained that allow for robust extraction of corresponding regions from a database of target images. The methods further include the use of sliding windows to decompose the query and target images into regions, and the clustering of the regions utilizing a novel similarity metric that ensures robust image matching in low response times.
Claims16
A method of imaging an object, comprising the steps of:
- decomposing a query image of the object into a set of plural regions;
- deriving a set of signatures for each region in the set of regions;
- defining a similarity measure between the query image and a target image as a function of a fraction of an area of the query and target images covered by matching pairs of regions from the set of regions of the query image and regions in the target image;
- and
- extracting the signatures of the set of regions of the query image by grouping in clusters the signatures of the regions having the similarity measure to identify whether the signatures for the regions have related pixel values to regions defined in the target image, to thereby image the object;
- wherein said step of decomposing the query image of the object into a set of plural regions comprises the steps of generating for each region of the query image a set of windows having varying sizes of pixel data between a minimum size of pixel data and a maximum size of pixel data, and clustering the generated windows such that ones of the windows that are within a certain distance of others of the windows are clustered together, to thereby define the plural regions in the query image of the object.
A method of matching query images with a target image stored in a library of target images, comprising the steps of:
- decomposing a query image into a plurality of regions, to thereby define a set of query regions;
- computing a set of signatures for each of the regions in the set of query regions;
- clustering the target images into a plurality of target image regions, to thereby define a set of target regions;
- computing a set of signatures for each of the regions in the set of target regions;
- computing a similarity measure to determine whether the query image and the target image can be matched by identifying a fraction of an area of the query image and of the target image that is covered by matching pairs of regions from the set of query regions and the set of target regions;
- and
- comparing the signatures of the query regions with the clustered signatures of the target regions to identify those signatures of the query regions that fall within the computed similarity measure with the clustered signatures of the target regions, and thereby match the query image with the target image;
- wherein said step of decomposing comprises generating for each query region a set of windows having varying sizes of pixel data between a minimum size of pixel data and a maximum size of pixel data, and clustering the windows such that those windows that are within a certain distance of others of the windows are clustered together, to thereby define the plural regions in the query image.
Dependencies are inferred from the claim text.