基于深度学习的图像色彩一致性算法设计  

Design of Image Color Consistency Algorithm Based on Deep Learning

作　　者：陈芳[1] CHEN Fang(Department of Computer Science and Technology,Suzhou College of Information Technology,Suzhou 215000,China)

机构地区：[1]苏州信息职业技术学院计算机科学与技术系,江苏苏州215000

出　　处：《自动化仪表》2023年第9期61-64,70,共5页Process Automation Instrumentation

基　　金：江苏省高等职业教育高水平专业群基金资助项目(2020-74)。

摘　　要：为了提高色彩一致性算法的精度和速度,提出1种基于卷积神经网络(CNN)的图像色彩一致性算法。所使用的CNN由2个卷积层、1个池化层和2个全连接层组成,以图像块为输入,而非之前常用的手工特征。此外,不同于之前的CNN算法,所提算法在局部区域进行,能够结合特征学习和回归形成一个端到端的优化过程。试验结果表明,所提算法在标准的原始RAW图像数据集上表现良好,且优于对比的基于假设和基于统计学习的算法。所提算法为使用CNN解决多光源问题提供了思路。To improve the accuracy and speed of the color consistency algorithm,an image color consistency algorithm based on convolutional neural network(CNN) is proposed.The CNN used consists of two convolutional layers,one pooling layer,and two fully connected layers,with image blocks as inputs instead of the manual features commonly used before.In addition,unlike previous CNN algorithms,the proposed algorithm is performed in a local region and is able to combine feature learning and regression to form an end-to-end optimization process.Experimental results show that the proposed algorithm achieves performing well on the standard original RAW image dataset and outperforms the compared hypothesis-based and statistical learning-based algorithms.The proposed algorithm provides an idea to solve the multiple light source problem using CNN.

关 键 词：卷积神经网络 特征学习 回归 端到端 色彩一致性 

分 类 号：TH164[机械工程—机械制造及自动化]