High reality image generation for DNN learning based on varying pixel intensity value model depend on each camera: the last 1％ accuracy improvement

Yusuke Kamiya; Nobuyuki Shinohara; Manabu Hashimoto

doi:10.1117/12.2521615

22 March 2019 High reality image generation for DNN learning based on varying pixel intensity value model depend on each camera: the last 1％ accuracy improvement

Yusuke Kamiya, Nobuyuki Shinohara, Manabu Hashimoto

Author Affiliations +

Proceedings Volume 11049, International Workshop on Advanced Image Technology (IWAIT) 2019; 1104945 (2019) https://doi.org/10.1117/12.2521615
Event: 2019 Joint International Workshop on Advanced Image Technology (IWAIT) and International Forum on Medical Imaging in Asia (IFMIA), 2019, Singapore, Singapore

Abstract

In object recognition using deep neural networks (DNNs) in the field of industry, the recognition accuracy rate decreases because of the differences in characteristics between the camera for learning and the camera for recognition. In this research, we solve this problem by statistically modeling the varying pixel intensity value of each recognition camera on the basis of actual acquired learning images. Here, the characteristics of generated images must be similar to images captured by the recognition camera. By using the statistical model, already-captured learning image sets can be converted to virtual images, which are accurately captured by the recognition camera. Through experiments using actual images, we confirmed that the recognition accuracy rate by our method is at least 1.0% higher than that of the conventional method based on Gaussian noise.

Citation Download Citation

Yusuke Kamiya, Nobuyuki Shinohara, and Manabu Hashimoto "High reality image generation for DNN learning based on varying pixel intensity value model depend on each camera: the last 1％ accuracy improvement", Proc. SPIE 11049, International Workshop on Advanced Image Technology (IWAIT) 2019, 1104945 (22 March 2019); https://doi.org/10.1117/12.2521615

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available