Sublinear time classification via feature padding and hashing

摘要

A linear function describing a framework for identifying an object of class k in an image sample x may be described by: wk*x+bk, where bk is the bias term. The higher the value obtained for a particular classifier, the better the match or strength of identity. A method is disclosed for classifier and/or content padding to convert dot-products to distances, applying a hashing and/or nearest neighbor technique on the resulting padded vectors, and preprocessing that may improve the hash entropy. A vector for an image, an audio, and/or a video may be received. One or more classifier vectors may be obtained. A padded image, video, and/or audio vector and classifier vector may be generated. A dot product may be approximated and a hashing and/or nearest neighbor technique may be performed on the approximated dot product to identify at least one class (or object) present in the image, video, and/or audio.

主权项

1. A computer-implemented method, comprising:
receiving an image vector that represents an input digital image; obtaining a plurality of classifier vectors, each classifier vector corresponding to at least one class; generating, by one or more computers, a padded image vector; generating, by the one or more computers, a plurality of padded classifier vectors comprising each of the plurality of classifier vectors padded with a scalar corresponding to the classifier vector; and classifying, by the one or more computers, the input digital image as including an object from at least one of the classes corresponding to the classifier vectors by approximating a dot product between the padded image vector and at least one of the plurality of padded classifier vectors.