Identifying features for media file comparison

摘要

Embodiments of the disclosed subject matter construct linear combinations of a set of defined “basic” features for each pair of sample and known media files. The linear combination of those features that give the optimal correlation coefficients (i.e., most-correlated) is then found. The highest correlation coefficients then indicate the “best” features to use in detecting a match of the sample file.

主权项

1. A computer-implemented method comprising:
selecting multiple different subsets of features of media content, from among a defined set of features of media content; generating, for each of the multiple different subsets of features, a respective linear combination of the features; for each linear combination of the features, comparing a sample item of media content to pre-existing items of media content, including items of media content that are labeled as matching the sample item of media content and items of media content that are labeled as not matching the sample item of media content, to generate a respective value for each feature of the linear combination, for each pre-existing item of media content that the sample item of media content is compared against; for each linear combination of the features, generating a correlation coefficient based on the respective value for each feature of the linear combination, wherein a correlation coefficient is a value that reflects a reliability of the linear combination in indicating a match or mismatch between a pair of items of media content; selecting a particular linear combination of the features based at least on the correlation coefficient for the particular linear combination of the features; and using the particular linear combination of the features in determining whether the another item of media content matches one or more of the pre-existing items of media content.