一种无边缘增强效应的羽毛图像光照校正算法  

Novel Feather Image's Illumination Normalization without Edge Enhancement Effect

作　　者：明俊峰[1] 汪仁煌[2] 唐苏湘[2] 岳洪伟[2] 

机构地区：[1]广东工业大学计算机学院,广州510006 [2]广东工业大学自动化学院,广州510006

出　　处：《科学技术与工程》2014年第22期245-249,共5页Science Technology and Engineering

基　　金：广东省自然科学基金重点项目(S2012020011071);广东省战略性新兴产业核心技术攻关项目(2012A010701004);广东省教育部产学研合作项目(2012B091000037;2012B091000041)资助

摘　　要：针对一般的图像光照校正算法会增强光照边缘特征,降低缺陷检测的准确率这一问题,提出了一种基于Retinex的自适应平滑改进算法。首先构造了一个新的传导函数,并且证明了传导函数的参数无论如何取值,算法不具有明暗边缘的增强效应,使得光照校正后羽毛纹理没有被增强;然后结合羽毛图像的特点,对局部不一致性的测量指标进行了改进。实验结果表明改进后的算法较好的满足了羽毛图像光照校正的需求,有效的提高缺陷检测的准确率。General illumination normalization algorithms can enhance the illumination edge features and reduce the accuracy of defect detection.An adaptive smoothing algorithm based on Retinex is proposed to alleviate this problem.First,a new conduction function is constructsed,and proved that the conduction function has no edge enhancement,whatever parameter values.So the feather texture will not be enhanced after normalizing illumination.Taking advantages of feather image characteristics,then,a measurement of local inhomogeneity was proposed.The experimental results show that the improved algorithm can normalize Illumination effectively,and improve the defect detection accuracy.

关 键 词：羽毛缺陷检测 光照校正 自适应平滑 传导函数 局部不一致性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]