The enhancement of depth estimation based on multi-scale convolution kernels

Heng Hua; Xinzhu Sang; Xiyu Tian; Wanqi Sun; Duo Chen; Peng Wang

doi:10.1117/12.2500775

7 November 2018 The enhancement of depth estimation based on multi-scale convolution kernels

Heng Hua, Xinzhu Sang, Xiyu Tian, Wanqi Sun, Duo Chen, Peng Wang

Proceedings Volume 10817, Optoelectronic Imaging and Multimedia Technology V; 1081710 (2018) https://doi.org/10.1117/12.2500775
Event: SPIE/COS Photonics Asia, 2018, Beijing, China

Abstract

Depth prediction is essential for three-dimensional optical displays. The accuracy of the depth map influences the quality of virtual viewpoint synthesis. Due to the relatively simple end-to-end structures of CNNs, the performance for poor and repetitive texture is barely satisfactory. In consideration of the shortage of existing network structures, the two main structures are proposed to optimize the depth map. (i) Inspired by GoogLeNet, the inception module is added at the beginning of the network. (ii) Assuming that the disparity map has only horizontal disparity, two sizes of rectangular convolution kernels are introduced to the network structure. Experimental results demonstrate that our structures of the CNN reduce the error rate from 19.23% to 14.08%.

Citation Download Citation

Heng Hua, Xinzhu Sang, Xiyu Tian, Wanqi Sun, Duo Chen, and Peng Wang "The enhancement of depth estimation based on multi-scale convolution kernels", Proc. SPIE 10817, Optoelectronic Imaging and Multimedia Technology V, 1081710 (7 November 2018); https://doi.org/10.1117/12.2500775

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