Robust audio hashing scheme based on cochleagram and cross recurrence analysis

A robust hashing method based on computer-vision techniques is proposed as an improvement to the existing computer-vision based hashing scheme. The cochleagram of the audio is treated as an image, from which Speeded Up Robust Features are extracted as essential features. Non-negative matrix factorisation is used to reduce the features´ dimension. In hashing matching, recurrence quantification analysis is performed on the cross recurrence plot that is constructed from the essentialfeatures of two clips to measure their similarity. Experimental results illustrate that the proposed method exhibits superior performance compared to existing techniques in identification rate (under various content preserving manipulations) and computational complexity.