Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1999-04-30
2001-02-27
Boudreau, Leo (Department: 2721)
Image analysis
Pattern recognition
Feature extraction
C382S210000, C382S236000, C382S278000, C382S280000, C359S559000
Reexamination Certificate
active
06195460
ABSTRACT:
TECHNICAL FIELD
The present invention relates to a pattern extraction apparatus for extracting moving patterns and differences between registration and collation patterns by collating N-dimensional patterns (e.g., voice patterns (one-dimensional), plane image patterns (two-dimensional), and stereoscopic patterns (three-dimensional) on the basis of spatial frequency characteristics.
BACKGROUND ART
Conventionally, a difference between two similar patterns has been extracted by a human visual check. More specifically, one of the patterns is set as a reference pattern, and the reference pattern and the other pattern are compared through the human eye, thereby extracting a difference.
In addition, common patterns (moving patterns) that are present at different positions in two similar patterns (overall patterns) are also extracted by a human visual check. More specifically, one of the patterns is set as a reference pattern, and the reference pattern and the other pattern are compared through the human eye, thereby extracting moving patterns.
A human visual check, however, can cope with only a case wherein a difference or moving pattern between two similar pattern is clear. That is, if overall patterns are complicated or a difference or moving pattern is small, it takes time to extract it, and an accurate check cannot be made. If a plurality of moving patterns are present, it is difficult to detect all the moving patterns.
It is an object of the present invention to provide a pattern extraction apparatus which can accurately extract a difference and moving pattern between similar patterns in a short period of time.
DISCLOSURE OF INVENTION
According to the present invention, registration Fourier N-dimensional pattern data is generated by performing N-dimensional discrete Fourier transform for N-dimensional pattern data of a registration pattern, collation Fourier N-dimensional pattern data is generated by performing N-dimensional discrete Fourier transform for N-dimensional pattern data of a collation pattern, one of the N-dimensional discrete Fourier transform and N-dimensional inverse discrete Fourier transform is performed in a first pattern processing means for synthesized Fourier N-dimensional pattern data obtained by synthesizing the registration Fourier N-dimensional pattern data and the collation Fourier N-dimensional pattern data, a correlation peak in a correlation component area appearing in the synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed is obtained, a portion around the obtained correlation peak is masked, the N-dimensional inverse discrete Fourier transform is performed for the synthesized Fourier N-dimensional pattern data in which the portion is masked when the N-dimensional discrete Fourier transform is performed in the first pattern processing means, the N-dimensional discrete Fourier transform is performed for the pattern data when the N-dimensional inverse discrete Fourier transform is performed in the first pattern processing means, and the N-dimensional inverse discrete Fourier transform is performed for re-synthesized Fourier N-dimensional pattern data generated by re-synthesizing the synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed and the registration Fourier N-dimensional pattern data.
According to the present invention, registration Fourier N-dimensional pattern data is generated by performing N-dimensional discrete Fourier transform for N-dimensional pattern data of a registration pattern, and collation Fourier N-dimensional pattern data is generated by performing N-dimensional discrete Fourier transform for N-dimensional pattern data of a collation pattern. The N-dimensional discrete Fourier transform or N-dimensional inverse discrete Fourier transform is performed in the first pattern processing means for synthesized Fourier N-dimensional pattern data obtained by synthesizing the registration Fourier N-dimensional pattern data and the collation Fourier N-dimensional pattern data. A correlation peak in a correlation component area appearing in the synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed is obtained. A portion around the obtained correlation peak is masked. The N-dimensional inverse discrete Fourier transform is performed for the synthesized Fourier N-dimensional pattern data in which the portion is masked when the N-dimensional discrete Fourier transform is performed in the first pattern processing means. The N-dimensional discrete Fourier transform is performed for the pattern data when the N-dimensional inverse discrete Fourier transform is performed in the first pattern processing means. The N-dimensional inverse discrete Fourier transform is performed for re-synthesized Fourier N-dimensional pattern data generated by re-synthesizing the synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed and the registration Fourier N-dimensional pattern data. The contour of a difference or moving pattern is extracted from the re-synthesized Fourier N-dimensional pattern data having undergone the inverse discrete Fourier transform. The location of the difference or moving pattern can be known whatsoever.
According to the present invention, registration Fourier N-dimensional pattern data and collation Fourier N-dimensional pattern data are synthesized, and amplitude suppression processing (log processing, root processing, or the like) is performed for the resultant synthesized Fourier N-dimensional pattern data. One of the N-dimensional discrete Fourier transform or N-dimensional inverse discrete Fourier transform is performed for the data, and amplitude restoration processing (inverse function processing of log processing, root processing, or the like) is performed for the synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed by the second pattern processing means. The synthesized Fourier N-dimensional pattern data having undergone the amplitude restoration processing and the registration Fourier N-dimensional pattern data are re-synthesized. N-dimensional inverse discrete Fourier transform is performed for the resultant re-synthesized Fourier N-dimensional pattern data.
In addition, according to the present invention, registration Fourier N-dimensional pattern data and collation Fourier N-dimensional pattern data are synthesized, and amplitude suppression processing (log processing, root processing, or the like) is performed for the resultant synthesized Fourier N-dimensional pattern data. One of the N-dimensional discrete Fourier transform or N-dimensional inverse discrete Fourier transform is performed for the data. The synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed by the second pattern processing means and the registration Fourier N-dimensional pattern data are re-synthesized. N-dimensional inverse discrete Fourier transform is performed for the resultant re-synthesized Fourier N-dimensional pattern data.
Furthermore, according to the present invention, registration Fourier N-dimensional pattern data is generated by performing amplitude suppression processing (log processing, root processing, or the like) for the N-dimensional pattern data of a registration pattern after performing N-dimensional discrete Fourier transform is performed for the pattern data. Collation Fourier N-dimensional pattern data is generated by performing amplitude suppression processing (log processing, root processing, or the like) for the N-dimensional pattern data of a collation pattern after performing N-dimensional discrete Fourier transform for the pattern data. The synthesized Fourier N-dimensional pattern data for which the Fourier transform has been performed by the second pattern processing means and the registration Fourier N-dimensional pattern data are re-synthesized. The N-dimensional inverse discrete Fourier transform is performed for the resultant re-synthesized Fourier N-dimen
Aoki Takafumi
Higuchi Tatsuo
Kawamata Masayuki
Kobayashi Koji
Nakajima Hiroshi
Blakely & Sokoloff, Taylor & Zafman
Boudreau Leo
Desire Gregory
Yamatake Corporation
LandOfFree
Pattern extraction apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Pattern extraction apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Pattern extraction apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2575015