Image recognition system based on cascaded over-complete dictionaries

A method for signal processing, comprising:nobtaining a plurality of signals related to a plurality of subjects;extracting, using at least one processor, one or more generic features of the subjects based on the signals;training, based on the one or more generic features, a first dictionary used to recognize the subjects, wherein the first dictionary is applied in an initial level of reconstruction of the signals;removing, using the at least one processor, the generic features from the plurality of signals to obtain one or more updated patterns for each subject;obtaining a set of discriminative features for each subject based on the updated patterns; andadding, based on the set of discriminative features, one or more second dictionaries used to recognize the subjects, wherein the one or more second dictionaries are applied at a level of reconstruction of the signals subsequent to the initial level.

An apparatus for signal processing, comprising:nmeans for obtaining a plurality of signals related to a plurality of subjects;means for extracting one or more generic features of the subjects based on the signals;means for training, based on the one or more generic features, a first dictionary to recognize the subjects, wherein the first dictionary is applied in an initial level of reconstruction of the signals;means for removing the generic features from the plurality of signals to obtain one or more updated patterns for each subject;means for obtaining a set of discriminative features for each subject based on the updated patterns; andmeans for adding, based on the set of discriminative features, one or more second dictionaries used to recognize the subjects, wherein the one or more second dictionaries are applied at a level of reconstruction of the signals subsequent to the initial level.

An apparatus for signal processing, comprising:na circuit configured to obtain a plurality of signals related to a plurality of subjects;at least one processor configured to:nextract one or more generic features of the subjects based on the signals;train, based on the one or more generic features, a first dictionary used to recognize the subjects, wherein the first dictionary is applied in an initial level of reconstruction of the signals;remove the generic features from the plurality of signals to obtain one or more updated patterns for each subject;obtain a set of discriminative features for each subject based on the updated patterns;add, based on the set of discriminative features, one or more second dictionaries used to recognize the subjects, wherein the one or more second dictionaries are applied at a level of reconstruction of the signals subsequent to the initial level; anda memory coupled with the at least one processor.

A computer-program product for signal processing comprising a non-transitory computer-readable storage medium having instructions stored thereon, the instructions executable by a processor for:nobtaining a plurality of signals related to a plurality of subjects;extracting, using at least one processor, one or more generic features of the subjects based on the signals;training, based on the one or more generic features, a first dictionary used to recognize the subjects, wherein the first dictionary is applied in an initial level of reconstruction of the signals;removing, using the at least one processor, the generic features from the plurality of signals to obtain one or more updated patterns for each subject;obtaining a set of discriminative features for each subject based on the updated patterns; andadding, based on the set of discriminative features, one or more second dictionaries used to recognize the subjects, wherein the one or more second dictionaries are applied at a level of reconstruction of the signals subsequent to the initial level.