基于BP神经网络的颜色补偿模型  

Color constancy model based on neural network

作　　者：丁淑芝 姜龙[2] 

机构地区：[1]淄博信息工程学校,山东淄博255038 [2]山东大学计算机学院,山东济南250101

出　　处：《计算机工程与设计》2010年第10期2294-2296,共3页Computer Engineering and Design

摘　　要：针对光源渐变等因素在机器视觉中产生的相关问题,提出了一种基于BP神经网络的图像颜色校正方法。该方法通过合适的训练集对BP神经网络进行大量训练,得到光照变化前后图像像素点之间的映射关系,从而建立了在渐变光照环境下的颜色恒常性模型。该方法不需要内建约束的自适应模型,对于输入的数据不需要对表面属性做特定假设,拥有自适应、自学习的特点。实验结果表明,该模型对室内真实环境中渐变日光下颜色的识别表现出较好的颜色恒常性。In order to solve the problem of color constancy of machine vision,a new method of color correction based on neural network is proposed.By means of appropriate sample sets,an improved back-propagation learning algorithm is adopted to train the neural network to obtain the mapping relation.After training,the image data with color constancy is obtained.The method does not need to build restricted adaptive models and does not need specific surface property assumptions.All the rules of color constancy are acquired in the process of neural network training.This model has a certain degree of robustness,self-adaption,and self-learning.At last,the method proved to be effective by testing.

关 键 词：颜色恒常性 机器视觉 神经网络 颜色较正 背景差 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]