A method, computer program, computer program product and system for representing visual information

A computer implemented method for representing a data set comprising at least one n dimensional data element (110) representing visual information, said method comprising the steps of:

• obtaining (210) said data set,
• obtaining (220) a dictionary ensemble (100) comprising a plurality of dictionaries (101) each comprising at least one basis function (102),
• assigning (230) each at least one data element (110) to a dictionary (101), wherein a set of basis functions (102) represents an m dimensional transformation domain,
• transforming (240) the at least one data element (110) with the corresponding dictionary (101) of basis functions (102) to the transformation domain wherein each data element (110) is defined by an associated coefficient set (120),
• sparsifying (250) the coefficient sets (120), and
• forming (260) the representation of the visual information comprising a coefficient data set (120) and the corresponding dictionaries (101),
• wherein:
• the obtained dictionary ensemble (100) comprises an ensemble of dictionaries formed from a combination of at least the at least one pre-cluster ensemble,
• each pre-cluster ensemble is obtained from a training data set of training visual information comprising training data elements,
• the training visual information is divided into pre-clusters,
• the pre-clusters are clustered by assigning each training data element (110) to a dictionary of basis functions (102) in a corresponding pre-cluster ensemble of dictionaries for each pre-cluster,
• each pre-cluster ensemble is trained to minimize the reconstruction error and maximize the sparseness of the training coefficient sets (120) of training data elements (110), and
• the clustering has been updated and the training iterated until at least one criteria is fulfilled.

A computer-implemented method (300) for learning an ensemble (100) of dictionaries, comprising:

• a step (310) of obtaining a training data set comprising training data elements representing n dimensional training visual information,
• a step (320) of dividing the training visual information into at least one pre-cluster,
• a step of creating a pre-cluster ensemble for each pre-cluster of training visual information, and
• a step (370) of combining at least the at least one pre-cluster ensemble to form an ensemble (100) of dictionaries,
• wherein the step of creating the at least one pre-cluster ensemble for each pre-cluster comprises
• initiating (330) a pre-cluster ensemble of dictionaries comprising basis functions for each pre-cluster,
• clustering (340) the at least one pre-cluster of training visual information by assigning each training data element to a dictionary of the corresponding pre-cluster ensemble,
• training (350) each pre-cluster ensemble to minimize the reconstruction error and maximize the sparseness of the training coefficient sets of training data elements, and
• updating (360) the clustering and iterating the training of the pre-cluster ensemble until at least one criteria is reached.