MACHINE LEARNING DEVICE AND CLASSIFICATION DEVICE FOR ACCURATELY CLASSIFYING INTO CATEGORY TO WHICH CONTENT BELONGS

• US 20160125273A1
• Filed: 09/10/2015
• Published: 05/05/2016
• Est. Priority Date: 10/31/2014
• Status: Active Grant

• Current Assignee: Casio Computer Company Limited
• Original Assignee: Casio Computer Company Limited
• Inventors: MATSUNAGA, Kazuhisa
• Granted Patent: US 9,659,239 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/2132 based on discrimination criteria, e.g. discriminant analysis
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06V 10/464 using a plurality of salient features, e.g. bag-of-words [BoW] representations
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods

Abstract

An image acquisition unit of a machine learning device acquires n learning images assigned with labels to be used for categorization (n is a natural number larger than or equal to 2). A feature vector acquisition unit acquires a feature vector representing a feature from each of the n learning images. A vector conversion unit converts the feature vector for each of the n learning images to a similarity feature vector based on a similarity degree between the learning images. A classification condition learning unit learns a classification condition for categorizing the n learning images, based on the similarity feature vector converted by the vector conversion unit and the label assigned to each of the n learning images. A classification unit categorizes unlabeled testing images in accordance with the classification condition learned by the classification condition learning unit.

Claims

1. A machine learning device comprising:

a content acquirer that acquires n learning contents (n is a natural number larger than or equal to

     2) with a label to be used for categorization;

a feature vector acquirer that acquires a feature vector from each of the n learning contents acquired by the content acquirer;

a vector converter that converts the feature vectors for the n learning contents acquired by the feature vector acquirer to similarity feature vectors based on similarity degrees between the learning contents;

a condition learning device that learns a classification condition for categorizing the n learning contents based on the similarity feature vectors converted by the vector converter and a label assigned to each of the n learning contents; and

a classifier that categorizes a testing content to which the label is not assigned, in accordance with the classification condition learned by the condition learning device.

2. The machine learning device according to claim 1, wherein the vector converter generates an n-row and n-column similarity matrix in which a number of the rows is n equal to the number n, a number of the columns is n equal to the number n, and each component of the similarity matrix is the similarity degree between the learning contents, and the vector converter converts elements of each row of the similarity matrix to elements of the similarity feature vector.

3. The machine learning device according to claim 1, wherein

the labels are for categorizing the n learning contents into two categories;

the condition learning device learns a function for linear separation as the classification condition in order to classify n points in a vector space determined by the similarity feature vectors of the n learning contents into two categories, based on the labels assigned to the n learning contents; and

the classifier classifies the testing contents into either one of the two categories in accordance with the function.

4. The machine learning device according to claim 3, further comprising:

an exponent generator that exponentiates the value of the similarity degree for each component of the similarity feature vector; and

a selector that selects a combination out of combinations of adjustment parameters and exponentiation parameters, the combination achieving a maximum correct answer percentage of the classification of the test contents by the classifier, the adjustment parameter defining a degree to which a learning error is allowed, the exponetiation parameter for changing the similarity degrees by exponetiation.

5. The machine learning device according to claim 3, wherein

the condition learning device learns the function for performing linear separation into the two categories by linear Support Vector Machine (linear SVM), k-nearest neighbor method, Naï

ve Bayes classification method, decision tree, neural network, bagging method, boosting method, or random forest method.

6. The machine learning device according to claim 1, wherein

the vector converter converts the feature vectors to similarity feature vectors with dimensions larger than dimensions of the feature vectors.

7. The machine learning device according to claim 1, wherein

the vector converter seeks the similarity degrees between the learning contents from histogram intersection.

8. The machine learning device according to claim 1, wherein

the feature vector acquirer acquires the feature vector in which a local feature is quantified from each of the n learning contents by BoF (Bag of Features) and normalizes, by L1 norm or L2 norm, the feature vector acquired from each of the n learning contents.

9. The machine learning device according to claim 1, wherein

each of the learning contents and the testing content comprises an image, a sound, or a text.

10. A classification device comprising:

an unknown-content classifier that categorizes an unknown content different from the learning contents and the testing content in accordance with a classification condition learned by the machine learning device according to claim 1.

11. The classification device according to claim 10, further comprising:

an acquirer that acquires from the unknown content a feature vector representing a feature of the unknown content; and

a converter that converts the feature vector of the unknown content acquired by the acquirer to a similarity feature vector based on a similarity degree between the learning contents and the unknown content, wherein

the unknown-content classifier categorizes the unknown content by classifying, in accordance with the classification condition, a point in a vector space determined from the similarity feature vector converted by the converter.

12. A machine learning method comprising:

a content acquisition step of acquiring n learning contents (n is a natural number larger than or equal to

     2) with a label to be used for categorization;

a feature vector acquisition step of acquiring a feature vector representing a feature from each of the n learning contents acquired by the content acquisition step;

a vector conversion step of converting the feature vectors for the n learning contents acquired by the feature vector acquisition step to a similarity feature vectors based on similarity degrees between the learning contents;

a learning step of learning a classification condition for categorizing the n learning contents based on the similarity feature vectors converted by the vector conversion step and a label assigned to each of the n learning contents; and

a classification step of categorizing a testing content to which the label is not assigned, in accordance with the classification condition learned by the learning step.

13. A classification method comprising:

an unknown-content classification step of categorizing an unknown content different from the learning contents and the testing content in accordance with a classification condition learned by the machine learning method according to claim 12.

14. A non-transitory computer readable recording medium recorded with a program for causing a computer of a machine learning device to implement:

a content acquirer that acquires n learning contents (n is a natural number larger than or equal to

     2) with a label to be used for categorization;

a feature vector acquirer that acquires a feature vector from each of the n learning contents acquired by the content acquirer;

a vector converter that converts the feature vectors for each of the n learning contents acquired by the feature vector acquirer to a similarity feature vectors based on similarity degrees between the learning contents;

a condition learning device that learns a classification condition for categorizing the n learning contents based on the similarity feature vectors converted by the vector converter and a label assigned to each of the n learning contents; and

a classifier that categorizes a testing content to which the label is not assigned, in accordance with the classification condition learned by the condition learning device.

15. A non-transitory computer readable recording medium recorded with a program for causing a computer of a classification device to implement

an unknown-content classifier that categorizes an unknown content different from the learning contents and the testing content in accordance with the classification condition learned through execution of the program set forth in claim 14.