Image recognition is of great significance currently, especially the one based on RGB images. Recently, as a substantial amount of RGB based CNNs showing great performance on image recognition problems, the initialization problem sees an increasingly importance as is highly associated with the variance as well as the convergence effect. Even though there exists several prevailing initialization methods, they didn’t consider the difference of contributions from different RGB channels on image recognition. Thus, in this paper, we put forward a proper approach to take the RGB contribution into account. Firstly, we conduct experiments to prove that RGB channels have respectively different contributions upon RGB based image recognition by the means that training one general CNN and three CNNs under datasets seperated from CIFAR-10 by RGB channel. Then, the ensemble of those CNNs shows how similar performance as the general one and also their respective accuracies are recorded as the influence proportion, namely their contribution ratio. As long as the influence proportion is obtained, according to that, we can initialize the weights of convolution filters in order to improve the effect of initialization.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
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.