US 12,190,237 B2
Pattern recognition device and learned model
Issei Nakamura, Chiyoda-ku (JP); Toshiaki Sugimura, Chiyoda-ku (JP); and Hayato Akatsuka, Chiyoda-ku (JP)
Assigned to NTT DOCOMO, INC., Chiyoda-ku (JP)
Appl. No. 17/267,890
Filed by NTT DOCOMO, INC., Chiyoda-ku (JP)
PCT Filed Jul. 12, 2019, PCT No. PCT/JP2019/027775
§ 371(c)(1), (2) Date Feb. 11, 2021,
PCT Pub. No. WO2020/070943, PCT Pub. Date Apr. 9, 2020.
Claims priority of application No. 2018-188160 (JP), filed on Oct. 3, 2018.
Prior Publication US 2021/0166063 A1, Jun. 3, 2021
Int. Cl. G06N 3/08 (2023.01); G06F 18/21 (2023.01); G06F 18/2113 (2023.01); G06N 20/00 (2019.01); G06V 10/56 (2022.01); G06V 10/774 (2022.01)
CPC G06N 3/08 (2013.01) [G06F 18/2113 (2023.01); G06F 18/217 (2023.01); G06N 20/00 (2019.01); G06V 10/56 (2022.01); G06V 10/774 (2022.01)] 6 Claims
OG exemplary drawing
 
1. A pattern recognition device comprising:
processing circuitry configured to
receive input data as image data;
perform pattern recognition to recognize an object included in the image data;
acquire a plurality of recognition candidates obtained by the pattern recognition and having respective recognition scores indicating certainty of recognition;
calculate an evaluation value obtained by quantifying a possibility of a second recognition candidate being correct answer data corresponding to the input data on the basis of a feature quantity regarding a first recognition candidate and the second recognition candidate, for each pair of the first recognition candidate that is a recognition candidate having a highest recognition score and the second recognition candidate that is another recognition candidate among the plurality of recognition candidates;
determine a final recognition result from among the plurality of recognition candidates on the basis of the evaluation value for each of the calculated second recognition candidates; and
output a pre-stored image corresponding to the final recognition result as the recognized object,
wherein the pattern recognition includes a first pattern recognition and a second pattern recognition different from the first pattern recognition,
the plurality of recognition candidates acquired including a recognition candidate obtained by the first pattern recognition for the input data and a recognition candidate obtained by the second pattern recognition for the input data,
the first pattern recognition is a scheme for recognizing an object included in the image data on the basis of a feature quantity extracted from a grayscale image obtained by performing grayscale conversion on the image data, and
the second pattern recognition is a scheme for recognizing the object included in the image data on the basis of a feature quantity including color information of the image data.