Basis learning for sparse image representation and classification and low data rate compression
AssigneeRockwell Collins, Inc.
InventorsShubha L. Kadambe, Lawrence Carin
Life of this patent
Estimating the term…
Abstract
A computerized method for transforming a signal representative of image information, the signal being a sparse signal, includes obtaining the signal. The method further includes using the signal to learn a dictionary of basis functions that represent at least part of the signal. The method further includes using the learned dictionary of basis functions to transform the signal.
Claims21
A computerized method for transforming a signal representative of image information, the signal being a sparse signal, the method comprising:
- obtaining the signal;
- using the signal to learn a dictionary of basis functions that represent at least part of the signal based on the signal, wherein learning the dictionary of basis functions includes estimating a representation of the signal and using parameters determined from the signal, wherein the dictionary is learned using a Beta process, wherein the basis functions represent at least part of the signal, wherein the parameters include at least one of a number of basis functions in the dictionary of basis functions;
- and
- using the learned dictionary of basis functions to transform the signal, wherein the transformed signal based on the learned dictionary of basis functions, wherein transforming the signal includes performing at least one of denoising, inpainting, and compressive sensing on the signal.
A computerized system for transforming a signal representative of image information, the signal being a sparse signal, the signal being obtained from at least one of a receiver, a memory and a camera, the system comprising:
- a dictionary learning module configured to learn a dictionary of basis functions that represent at least part of the signal based on the signal;
- and
- a signal analysis module configured to use the learned dictionary of basis functions to transform the signal, wherein the signal is analyzed to determine parameters for learning the dictionary of basis functions for the signal, wherein the basis functions represent at least part of the signal, wherein the parameters include at least one of a number of basis functions in the dictionary of basis functions and a sparsity level of the signal, wherein the dictionary is learned using a Beta process;
- and
- wherein the signal is transformed based on the learned dictionary of basis functions, wherein the signal is at least one of denoised, inpainted, and compressive sensed.
A computerized method of transforming a signal representative of image information, the signal being a sparse signal, the method comprising:
- obtaining the signal;
- analyzing the signal to determine parameters for learning a dictionary of basis functions for the signal, wherein the basis functions represent at least part of the signal, wherein the parameters include at least one of a number of basis functions in the dictionary of basis functions and a sparsity level of the signal;
- estimating a representation of the signal;
- calculating an error between the signal and the representation of the signal;
- adjusting the parameters based on the calculated error;
- re-estimating the representation based on the adjusted parameters;
- repeating the calculating, adjusting and re-estimating until the error is below an error threshold;
- learning the dictionary of basis functions based on the determined parameters, wherein the dictionary is learned using a Beta process;
- and
- transforming the signal based on the learned dictionary of basis functions, wherein transforming the signal includes performing at least one of denoising, inpainting, and compressive sensing on the signal.
Dependencies are inferred from the claim text.