Approaching the computational color constancy as a classification problem through deep learning

作者：

Highlights：

• A deep learning framework for the computational color constancy problem is proposed.

• Illumination is estimated by solving an illumination classification problem.

• Convolutional neural network is trained to solve the classification problem.

• The proposed method outperforms previous methods on multiple test datasets.

摘要

Highlights•A deep learning framework for the computational color constancy problem is proposed.•Illumination is estimated by solving an illumination classification problem.•Convolutional neural network is trained to solve the classification problem.•The proposed method outperforms previous methods on multiple test datasets.

论文关键词：Computational color constancy,White balancing,Illumination estimation,Machine learning,Convolutional neural network

论文评审过程：Received 4 January 2016, Revised 7 July 2016, Accepted 15 August 2016, Available online 17 August 2016, Version of Record 26 August 2016.

论文官网地址：https://doi.org/10.1016/j.patcog.2016.08.013