Signal compression method, device, program, and recording medium; and signal retrieval method, device, program, and recording medium

A signal compression device which compresses an original signal which is provided in advance to convert the original signal into a compressed signal, comprising:

• an initial sub-signal creation section which creates, from the original signal, sub-signals of shorter length than the original signal;
• a created sub-signal selection section which, for each of the sub-signals which has been produced by the initial sub-signal creation section, prunes the created sub-signal candidates to those for which the amount of data is less than for the original signal;
• a sub-signal re-creation section which, using the created sub-signal candidates which have been produced by the created sub-signal selection section, determines upon a created sub-signal which is actually to be used;
• a compression mapping determination section which determines upon a mapping for calculation of a compressed signal from the respective sub-signals which has been obtained by the sub-signal re-creation section;
• and
• a signal compression section which calculates a compressed signal which corresponds to each of the sub-signals which have been obtained by the sub-signal re-creation section, based upon the mapping which has been obtained by the compression mapping determination section.

A signal compression program which compresses an original signal which has been provided in advance to convert the original signal into a compressed signal, for causing a computer to execute:

• an initial sub-signal creation step of creating, from the original signal, sub-signals of shorter length than the original signal;
• a created sub-signal selection step of, for each of the sub-signals which have been produced by the initial sub-signal creation step, pruning the created sub-signal candidates to those for which the amount of data is less than for the original signal;
• a sub-signal re-creation step of determining upon a created sub-signal which is actually to be used, using the created sub-signal candidates which have been produced by the created sub-signal selection step;
• a compression mapping determination step of determining, from each of the sub-signals which have been produced by the sub-signal re-creation step, a mapping for calculation of a compressed signal;
• and
• a signal compression step of calculating a compressed signal which corresponds to each of the sub-signals which have been obtained by the sub-signal re-creation step, based upon the mapping which has been obtained by the compression mapping determination step.

A recording medium capable of being read by a computer, upon which is recorded a signal compression program which compresses an original signal which has been provided in advance to convert the original signal into a compressed signal, for causing a computer to execute:

• an initial sub-signal creation step of creating, from the original signal, sub-signals of shorter length than the original signal;
• a created sub-signal selection step of, for each of the sub-signals which have been produced by the initial sub-signal creation step, pruning the created sub-signal candidates to those for which the amount of data is less than for the original signal;
• a sub-signal re-creation step of determining upon a created sub-signal which is actually to be used, using the created sub-signal candidates which have been produced by the created sub-signal selection step;
• a compression mapping determination step of determining, from each of the sub-signals which have been produced by the sub-signal re-creation step, a mapping for calculation of a compressed signal;
• and
• a signal compression step of calculating a compressed signal which corresponds to each of the sub-signals which have been obtained by the sub-signal re-creation step, based upon the mapping which has been obtained by the compression mapping determination step.

A signal retrieval method which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object, comprising:

• a query feature extraction step in which a feature is produced from the query signal;
• a database feature extraction step in which a window upon which attention is focused is set within the database signal, and in which a feature is produced from the database signal within the window upon which attention is focused;
• a database feature partitioning step in which a feature sequence which has been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused is partitioned;
• a database feature pruning step in which a representative feature is extracted from the feature sequence which has been obtained after partitioning by the database feature partitioning step, and a representative feature sequence is produced which consists of a smaller number of features;
• a feature region extraction step in which a region is produced in which a feature which is included in the partition which has been produced by the database feature partitioning step is present;
• a feature matching step in which a distance is calculated between a feature sequence which has been produced by the query feature extraction step and a representative feature sequence which has been produced by the database feature pruning step;
• a distance compensation step in which the distance which has been calculated by the feature matching step is compensated using the region which has been produced by the feature region extraction step;
• and
• a signal detection decision step in which, by comparing together the distance which has been produced after compensation by the distance compensation step and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal, and
• wherein the processing of the feature matching step through the signal detection decision step is repeated while shifting the window upon which attention is focused, for some locations within the database signal, the distance from the query signal is calculated, and it is determined whether or not the query signal is present at the locations within the database signal.

A signal retrieval method which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object, comprising:

• a query feature extraction step in which a feature is produced from a query signal;
• a database feature extraction step in which a window upon which attention is focused is set within the database signal, and in which a feature is produced from the database signal within the window upon which attention is focused;
• a database feature classification step in which the respective features which have been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused are classified based upon a distance which has been defined in advance, and a representative feature of the classification is determined upon;
• a selection threshold setting step in which a selection threshold for the distance which has been defined by the database feature classification step is calculated from a search threshold which has been defined in advance;
• a database feature selection step in which, among the classification which has been produced by the database feature classification step, a feature is selected which is included in the classification which contains a representative feature such that the distance from the feature which has been produced by the query feature extraction step satisfies a condition which is produced from the selection threshold which has been calculated by the selection threshold setting step;
• a segment extraction step in which segments, which are sub-sequences, are extracted by segmenting a feature sequence which has been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused;
• a compression mapping determination step in which, from each of the segments which have been obtained by the segment extraction step, a mapping is determined for calculation of a feature of less dimensions than the feature;
• a database feature compression step in which a feature which corresponds to a segment which has been obtained by the segment extraction step and which is of less dimensions than the feature is calculated based upon the mapping which has been obtained by the compression mapping determination step;
• a query feature compression step in which a feature which corresponds to a feature which has been obtained by the query feature extraction step and which is of less dimensions than the feature is calculated based upon the mapping which has been obtained by the compression mapping determination step;
• a feature matching step in which a distance is calculated between a compressed feature sequence which has been produced by the database feature compression step and a compressed feature which has been produced by the query feature extraction step;
• and
• a signal detection decision step in which, by comparing together the distance which has been calculated by the feature matching step and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal, and
• wherein the processing of the feature matching step and the signal detection decision step is repeated while shifting the window upon which attention is focused, for some locations within the database signal, the distance from the query signal is calculated;
• and it is determined whether or not the query signal is present at the locations within the database signal.

A signal retrieval device which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object, comprising:

• a query feature extraction section which produces a feature from the query signal;
• a database feature extraction section which sets a window upon which attention is focused is set within the database signal, and which produces a feature from the database signal within the window upon which attention is focused;
• a database feature partitioning section which partitions a feature sequence which has been produced by repeatedly performing the processing of the database feature extraction section while shifting the window upon which attention is focused;
• a database feature pruning section which extracts a representative feature from the feature sequence which has been obtained after partitioning by the database feature partitioning section, and which produces a representative feature sequence which consists of a smaller number of features;
• a feature region extraction section which produces a region in which a feature which is included in the partition which has been produced by the database feature partitioning section is present;
• a feature matching section which calculates a distance between a feature sequence which has been produced by the query feature extraction section and a representative feature sequence which has been produced by the database feature pruning section;
• a distance compensation section in which the distance which has been calculated by the feature matching section is compensated using the region which has been produced by the feature region extraction section;
• and
• a signal detection decision section in which, by comparing together the distance which has been produced after compensation by the distance compensation section and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal, and
• wherein the processing of the feature matching section through the signal detection decision section is repeated while shifting the window upon which attention is focused;
• for some locations within the database signal, the distance from the query signal is calculated;
• and it is determined whether or not the query signal is present at the locations within the database signal.

A signal retrieval device which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object, comprising:

• a query feature extraction section which produces a feature from a query signal;
• a database feature extraction section which sets a window upon which attention is focused within the database signal, and which produces a feature from the database signal within the window upon which attention is focused;
• a database feature classification section which classifies the respective features which have been produced by repeatedly performing the processing of the database feature extraction section while shifting the window upon which attention is focused based upon a distance which has been determined in advance, and determines upon a representative feature of the classification;
• a selection threshold setting section which calculates a selection threshold for the distance which has been defined by the database feature classification section from a search threshold which has been defined in advance;
• a database feature selection section which, among the classification which has been produced by the database feature classification section, selects a feature which is included in the classification which contains a representative feature such that the distance from the feature which has been produced by the query feature extraction section satisfies a condition which is produced from the selection threshold which has been calculated by the selection threshold setting section;
• a segment extraction section which extracts segments, which are sub-sequences, by segmenting a feature sequence which has been produced by repeatedly performing the processing of the database feature extraction section while shifting the window upon which attention is focused;
• a compression mapping determination section which, from each of the segments which have been obtained by the segment extraction section, determines a mapping for calculation of a feature of less dimensions than the feature;
• a database feature compression section which calculates a feature which corresponds to a segment which has been obtained by the segment extraction section and which is of less dimensions than the feature based upon a mapping which has been obtained by the compression mapping determination section;
• a query feature compression section which calculates a feature which corresponds to a feature which has been obtained by the query feature extraction section and which is of less dimensions than the feature based upon the mapping which has been obtained by the compression mapping determination section;
• a feature matching section which calculates a distance between a compressed feature sequence which has been produced by the database feature compression section and a compressed feature which has been produced by the query feature extraction section;
• and
• a signal detection decision section which, by comparing together the distance which has been calculated by the feature matching section and a search threshold, which is a threshold which corresponds to the distance, decides whether or not the query signal is present at the location within the database signal, and
• wherein the processing of the feature matching section through the signal detection decision section is repeated while shifting the window upon which attention is focused, for some locations within the database signal, the distance from the query signal is calculated;
• and it is determined whether or not the query signal is present at the locations within the database signal.

A program for execution by a computer of a signal retrieval device which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object, comprising:

• a query feature extraction step in which a feature is produced from the query signal;
• a database feature extraction step in which a window upon which attention is focused is set within the database signal, and in which a feature is produced from the database signal within the window upon which attention is focused;
• a database feature partitioning step in which a feature sequence which has been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused is partitioned;
• a database feature pruning step in which a representative feature is extracted from the feature sequence which has been obtained after partitioning by the database feature partitioning step, and a representative feature sequence is produced which consists of a smaller number of features;
• a feature region extraction step in which a region is produced in which a feature which is included in the partition which has been produced by the database feature partitioning step is present;
• a feature matching step in which a distance is calculated between a feature which has been produced by the query feature extraction step and a representative feature sequence which has been produced by the database feature pruning step;
• a distance compensation step in which the distance which has been calculated by the feature matching step is compensated using the region which has been produced by the feature region extraction step;
• a signal detection decision step in which, by comparing together the distance which has been produced after compensation by the distance compensation step and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal;
• and
• a step of:
• repeating the processing of the feature matching step through the signal detection decision step while shifting the window upon which attention is focused, for some locations within the database signal, calculating the distance from the query signal;
• and determining whether or not the query signal is present at the locations within the database signal.

A program for execution by a computer of a signal retrieval device which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object, comprising:

• a query feature extraction step in which a feature is produced from the query signal;
• a database feature extraction step in which a window upon which attention is focused is set within the database signal, and in which a feature is produced from the database signal within the window upon which attention is focused;
• a database feature classification step in which the respective features which have been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused are classified based upon a distance which has been determined in advance, and a representative feature of the classification is determined upon;
• a selection threshold setting step in which a selection threshold for the distance which has been defined by the database feature classification step is calculated from a search threshold which has been defined in advance;
• a database feature selection step in which, among the classification which has been produced by the database feature classification step, a feature is selected which is included in the classification which contains a representative feature such that the distance from the feature which has been produced by the query feature extraction step satisfies a condition which is produced from the selection threshold which has been calculated by the selection threshold setting step;
• a segment extraction step in which segments, which are sub-sequences, are extracted by segmenting a feature sequence which has been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused;
• a compression mapping determination step in which, from each of the segments which have been obtained by the segment extraction step, a mapping is determined for calculation of a feature of less dimensions than the feature;
• a database feature compression step in which a feature which corresponds to a segment which has been obtained by the segment extraction step and which is of less dimensions than the feature is calculated based upon the mapping which has been obtained by the compression mapping determination step;
• a query feature compression step in which a feature which corresponds to a feature which has been obtained by the query feature extraction step and which is of less dimensions than the feature is calculated based upon the mapping which has been obtained by the compression mapping determination step;
• a feature matching step in which a distance is calculated between a compressed feature sequence which has been produced by the database feature compression step and a compressed feature which has been produced by the query feature extraction step;
• a signal detection decision step in which, by comparing together the distance which has been calculated by the feature matching step and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal;
• and
• a step in which:
• the processing of the feature matching step and the signal detection decision step is repeated while shifting the window upon which attention is focused;
• for some locations within the database signal, the distance from the query signal is calculated;
• and it is determined whether or not the query signal is present at the locations within the database signal.

A recording medium capable of being read by a computer, upon which is recorded a program for causing a computer of a signal retrieval device which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object to execute:

• a query feature extraction step in which a feature is produced from the query signal;
• a database feature extraction step in which a window upon which attention is focused is set within the database signal, and in which a feature is produced from the database signal within the window upon which attention is focused;
• a database feature partitioning step in which a feature sequence which has been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused is partitioned;
• a database feature pruning step in which a representative feature is extracted from the feature sequence which has been obtained after partitioning by the database feature partitioning step, and a representative feature sequence is produced which consists of a smaller number of features;
• a feature region extraction step in which a region is produced in which a feature which is included in the partition which has been produced by the database feature partitioning step is present;
• a feature matching step in which a distance is calculated between a feature which has been produced by the query feature extraction step and a representative feature sequence which has been produced by the database feature pruning step;
• a distance compensation step in which the distance which has been calculated by the feature matching step is compensated using the region which has been produced by the feature region extraction step;
• a signal detection decision step in which, by comparing together the distance which has been produced after compensation by the distance compensation step and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal;
• and
• a step of:
• repeating the processing of the feature matching step through the signal detection decision step while shifting the window upon which attention is focused;
• for some locations within the database signal, calculating the distance from the query signal;
• and determining whether or not the query signal is present at the locations within the database signal.

A recording medium capable of being read by a computer, upon which is recorded a program for causing a computer of a signal retrieval device which finds out portions from a database signal which has been registered in advance which are similar to a query signal which is taken as an object to execute:

• a query feature extraction step in which a feature is produced from the query signal;
• a database feature extraction step in which a window upon which attention is focused is set within the database signal, and in which a feature is produced from the database signal within the window upon which attention is focused;
• a database feature classification step in which the respective features which have been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused are classified based upon a distance which has been determined in advance, and a representative feature of the classification is determined upon;
• a selection threshold setting step in which a selection threshold for the distance which has been defined by the database feature classification step is calculated from a search threshold which has been defined in advance;
• a database feature selection step in which, among the classification which has been produced by the database feature classification step, a feature is selected which is included in the classification which contains a representative feature such that the distance from the feature which has been produced by the query feature extraction step satisfies a condition which is produced from the selection threshold which has been calculated by the selection threshold setting step;
• a segment extraction step in which segments, which are sub-sequences, are extracted by segmenting a feature sequence which has been produced by repeatedly performing the database feature extraction step while shifting the window upon which attention is focused;
• a compression mapping determination step in which, from each of the segments which have been obtained by the segment extraction step, a mapping is determined for calculation of a feature of less dimensions than the feature;
• a database feature compression step in which a feature which corresponds to the segment which has been obtained by the segment extraction step and which is of less dimensions than the feature is calculated based upon the mapping which has been obtained by the compression mapping determination step;
• a query feature compression step in which a feature which corresponds to a feature which has been obtained by the query feature extraction step and which is of less dimensions than the feature is calculated based upon the mapping which has been obtained by the compression mapping determination step;
• a feature matching step in which a distance is calculated between a compressed feature sequence which has been produced by the database feature compression step and a compressed feature which has been produced by the query feature extraction step;
• a signal detection decision step in which, by comparing together the distance which has been calculated by the feature matching step and a search threshold, which is a threshold which corresponds to the distance, it is decided whether or not the query signal is present at the location within the database signal;
• and
• a step in which:
• the processing of the feature matching step and the signal detection decision step is repeated while shifting the window upon which attention is focused;
• for some locations within the database signal, the distance from the query signal is calculated;
• and it is determined whether or not the query signal is present at the locations within the database signal.