Method for reducing blocking artifacts in images
AssigneeMitsubishi Electric Research Laboratories, Inc.
InventorsFatih Porikli, Yi Wang
Life of this patent
Estimating the term…
Abstract
Blocking artifacts are reduced by projecting each patch obtained from an input image onto a set of bases vectors to determine multiple representations for each patch. The set of bases vectors are learned from a training image, and the bases vectors include a full basis vector, and one or two subspace bases vectors. An optimal basis vector is determined in the set of bases vectors for each patch according to the projection. A threshold is applied to coefficients of the optimal basis vector to determine a filtered representation for each patch, and a reconstructed patch is generated using the filtered representation. Then, the aggregating the reconstructed patches are aggregated to produce an output image.
Claims15
A method for reducing blocking artifacts, comprising:
- projecting each patch obtained from an input image onto a set of bases vectors to determine multiple representations for each patch, wherein the set of bases vectors are learned from a training image, and the bases vectors include a full basis vector, and one or two subspace bases vectors, wherein the training image is a step image;
- determining an optimal basis vector in the set of bases vectors for each patch according to the projection;
- applying a threshold to coefficients of the optimal basis vector to determine a filtered representation for each patch;
- generating a reconstructed patch using the filtered representation;
- and
- aggregating the reconstructed patches to produce an output image, wherein the steps are performed in a processor, further comprising:
- (a) initializing the set of bases vectors as a current set of bases vectors with the patches obtained from the training image;
- (b) determining a representation of each patch using the current set of basis vectors;
- (c) computing an error matrix for each current basis vector by using all other current basis vectors in the set of current bases vectors;
- (d) decomposing the error matrix only for the patches that use the set of current basis vector to obtain a truncated error matrix;
- (e) computing a largest eigenvalue and a corresponding eigenvector of the truncated error matrix;
- (f) assigning the eigenvector as an updated basis vector to be used as the current basis vector in a next iteration;
- and
- iterating the steps (a) through (f) for each basis vectors until a termination condition is reached.
Dependencies are inferred from the claim text.