Authors:
Minquan Wang
and
Zhaowei Shang
Affiliation:
College of Computer Science, Chongqing University, No. 174 Shazheng Street, Chongqing and China
Keyword(s):
Illumination Estimation, Deep Convolution Network, Separable Convolution, Global Average Pooling.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Computational Intelligence
;
Health Engineering and Technology Applications
;
Human-Computer Interaction
;
Methodologies and Methods
;
Neural Networks
;
Neurocomputing
;
Neurotechnology, Electronics and Informatics
;
Pattern Recognition
;
Physiological Computing Systems
;
Sensor Networks
;
Signal Processing
;
Soft Computing
;
Theory and Methods
Abstract:
Illumination estimation has been studied for a long time. The algorithms to solve the problem can be roughly divided into two categories: statistical-based and learning-based. Statistical-based algorithm has the advantage of fast computing speed but low accuracy. Learning-based algorithm improve the estimation accuracy to some extent, but generally have high computational complexity and storage space. In this paper, a new deep convolution neural network is proposed. We design the network with more layers (11 convolution layers) than the existing methods, remove the “skip connection” and “Global Average Pooling” is used to replace “Fully Connection” layer which is commonly used in the existing methods. We use the separable convolution instead of the standard convolution to reduce the number of parameters. In reprocessed Color Checker Dataset, compared with the present state-of-the-art the proposed method reduces the average angular error by about 60%. At the same time, using separable
convolution and “Global Average Pooling” reduces the number of parameters by about 86% compared with do not use them.
(More)