Classification of images using a dictionary of compressed time-frequency atoms

A method for classifying unknown test images based on their similarities with a dictionary of example target and non-target images, comprising the steps of:

• receiving one of the unknown test images;
• identifying a scaling coefficient, wherein the scaling coefficient is set to a maximum value produced by taking the inner product of the unknown test image and each one of the example target and non-target images in the dictionary
• scaling the dictionary by the scaling coefficient;
• substituting a difference between the unknown test image and the scaled dictionary as the unknown test image;
• repeating the steps of identifying a scaling coefficient, scaling and substituting until a predetermined number of iterations have been performed;
• and
• classifying the test image by separately summing the scaling coefficients associated with the target and the non-target images in the dictionary and choosing a classification corresponding to the greatest sum.

A method for classifying unknown test images based on their similarities with a dictionary of example target and non-target images, comprising the steps of:

• receiving one of the unknown test images;
• projecting the test image onto each of the images in the dictionary of example target and non-target images;
• identifying a scaling coefficient, wherein the scaling coefficient is set to a maximum value produced by the projection of the test image onto each one of the example target and non-target images in the dictionary and is associated with the image identified as producing the maximum value;
• generating a residue of the test image;
• refining a linear expansion of scaled dictionary images;
• repeating, using the residue as the test image, the steps of projecting the test image onto the dictionary, identifying a scaling coefficient, generating a residue and refining the linear expansion until a predetermined number of iterations have been performed;
• and
• classifying the test image by comparing the sum of the scaling coefficients associated with the target and non-target images in the dictionary.