Presentation | 2023-02-02 Recognition of imperfect image based on deep learning Takuto Nakazawa, Shigeru Kubota, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Even though the visual information obtained is imperfect, we can estimate a correct answer by using a knowledge that we already possess. For example, in cases where we see a three-dimensional structure of blocks that contains some hidden ones, we can estimate the total number of blocks from the mechanistic knowledge (e.g., the existence of an upper block means the existence of lower blocks, which mechanistically support the upper one). In this study, we have addressed an problem to estimate the total number of blocks in a 3-dimensional block structure. The visual images of block structure, which includes 11-20 blocks, were constructed using a PC, and they were learned by a convolutional neural network through the deep learning algorithm. The result showed that the network can predict the total number of blocks with a high accuracy of ?90%. This finding indicates a possibility that the deep learning network can be used as a mathematical model for human cognitive recognition. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Imperfect information / Deep learning / Convolutional neural network / Image classification |
Paper # | IE2022-51 |
Date of Issue | 2023-01-26 (IE) |
Conference Information | |
Committee | IE |
---|---|
Conference Date | 2023/2/2(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | NII |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Image Processing |
Chair | Kazuya Kodama(NII) |
Vice Chair | Hiroyuki Bandoh(NTT) / Toshihiko Yamazaki(Univ. of Tokyo) |
Secretary | Hiroyuki Bandoh(KDDI Research) / Toshihiko Yamazaki(Nagoya Inst. of Tech.) |
Assistant | Shunsuke Iwamura(NHK) / Shinobu Kudo(NTT) |
Paper Information | |
Registration To | Technical Committee on Image Engineering |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Recognition of imperfect image based on deep learning |
Sub Title (in English) | Estimating the number of 3-dimentional building blocks |
Keyword(1) | Imperfect information |
Keyword(2) | Deep learning |
Keyword(3) | Convolutional neural network |
Keyword(4) | Image classification |
1st Author's Name | Takuto Nakazawa |
1st Author's Affiliation | Yamagata University(Yamagata Univ.) |
2nd Author's Name | Shigeru Kubota |
2nd Author's Affiliation | Yamagata University(Yamagata Univ.) |
Date | 2023-02-02 |
Paper # | IE2022-51 |
Volume (vol) | vol.122 |
Number (no) | IE-376 |
Page | pp.pp.1-5(IE), |
#Pages | 5 |
Date of Issue | 2023-01-26 (IE) |