Method for signal encoding and feature extraction
AssigneeLucent Technologies Inc.
InventorsDaniel D. Lee, Hyunjune Sebastian Seung
Life of this patent
Estimating the term…
Abstract
A new encoder is provided which operates on a database of n-dimensional signal vectors where such signal vectors are restricted to being non-negative. In the establishment of a codebook for this encoder, p signal vectors (of dimension n) are placed in an n.times.p matrix X, and an approximate factorization: X.apprxeq.WV is constructed, where W and V are matrices whose entries are non-negative. Each column of the n.times.r matrix W is characterized as a "feature", such features having been "learned" from the database of n-dimensional signal vectors.
Claims16
A method for encoding an input signal comprising the steps of:
- operating on a database of signal values to provide a codebook of features representative of a general class of signals included in said signal values, wherein said codebook features are characterized as discrete portions of signals comprising said database;
- and
- encoding an unknown input signal in respect to said features of said codebook by causing said input signal to be expressed in terms of a plurality of coefficient terms and a related plurality of said codebook features, wherein said encoded input signal is expressed as a combination of said discrete portions.
A system for encoding an input signal comprising:
- processor means for operating on database of signal values to provide a codebook of features representative of a general class of signals included in said signal values, wherein said codebook features are characterized as discrete portions of signals comprising said database;
- encoding means for encoding an unknown input signal in respect to said features of said codebook by causing said input signal to be expressed in terms of a plurality of coefficient terms and a related plurality of said codebook features, wherein said encoded input signal is expressed as a combination of said portions of said signals.
In a pattern recognition system wherein an input signal representing parameters of an unknown pattern of interest is encoded relative to a predetermined codebook and thereafter a set of decision rules is applied to said encoded signal for finding a closest match between said unknown pattern and a one of a database of known patterns, an improved method for determining said codebook and for carrying out said encoding of said input signal comprising the steps of:
- operating on said database of known patterns to learn a codebook of features representative of a class of patterns included in said database, wherein said codebook features are characterized as discrete portions of patterns comprising said database;
- and
- encoding said input signal in respect to said features of said codebook by causing said input signal to be expressed in terms of a plurality of coefficient terms and a related plurality of said codebook features, wherein said encoded input signal is expressed as a combination of said portions of said signals.
An encoder comprising:
- learning means operating on database of training signal values to learn a codebook of features representative of a general class of signals included in said training signal values, wherein said codebook features are characterized as discrete portions of signals comprising said database;
- encoding means for encoding an unknown input signal in respect to said features of said codebook by causing said input signal to be expressed in terms of a plurality of coefficient terms and a related plurality of said codebook features, wherein said encoded input signal is expressed as a combination of said portions of said signals.
Dependencies are inferred from the claim text.