Multi-category recognizer generation device and method thereof, data recognition device and method thereof

Abstract

The feature represented by the multi-class classifier recognition data belongs to which class among the plurality of classes, comprising: a first class classifier generation unit that generates a plurality of first class classifiers for 1-to-N recognition; and a second class classifier a layer recognizer generating unit that generates a second layer feature vector by concatenating the score values ​​respectively outputted by a plurality of first layer recognizers, and generates a multi-level feature vector for 1-to-N recognition by inputting the second layer feature vector; a second-level classifier, the multi-class classifier will generate a second-level feature vector by concatenating the score values ​​output by the plurality of first-level classifiers by inputting data to a plurality of first-class classifiers, The classification corresponding to the second-level classifier that outputs the largest score value by inputting the second-level feature vector to a plurality of second-level classifiers is recognized as the class to which the feature represented by the input data belongs. Classification.

Description

[0001] This application claims the priority of the Japanese patent application No. 2012-236440 for which it applied on October 26, 2012 as a basic application, and takes in all the content of this basic application in this application. technical field [0002] The invention relates to a multi-category identifier, a data identification device, a multi-category identification method and a data identification method. Background technique [0003] There are cases where you want to know the name of a flower found in the mountains or on the roadside. For this reason, the following technology is proposed (for example, referring to Japanese Unexamined Patent Application Publication No. 2002-203242): by using digital images of flowers and leaves obtained by photography, etc., using a clustering method to extract a large number of local features of flowers and leaves as objects, Furthermore, a single or a plurality of feature quantities using the histogramized information of the extra...

AI Technical Summary

Problems solved by technology

In this way, when a multi-class classifier is constructed by combining two class classifiers, there is a problem that the classification performance of each class becomes uneven, that is, the normalization is not performed.

[0006] Furthermore, when a multi-class classifier is formed by combining 2-class classifiers, if a sufficient amount of learning data is not given to each 2-class classifier, the recognition performance of each class will become uneven.

However, for example, in the classification of flower types, there is a problem that it is difficult to provide a sufficient amount of learning data for all classifications.

Images