基于bagging-rough SVM集成的去马赛克方法

来源期刊:中南大学学报(自然科学版)2015年第6期

论文作者:赵佰亭 贾晓芬 周孟然 陈兆权 黄贤波

文章页码:2065 - 2074

关键词:去马赛克;支持向量机;集成;粗糙集

Key words:demosaicing; support vector machine; ensemble; rough set

摘    要:能否有效保护图像的细节信息是衡量去马赛克技术优劣的关键因素,为改善图像细小边缘区域的边缘特征,抑制伪彩色效应或锯齿现象,提出一种基于bagging-rough的SVM集成算法,并利用该算法实现去马赛克。为提高支持向量机集成的预测精度,利用各平面之间的色彩相关性及色差恒定原理构建色差平面,在色差平面上利用图像的空间相关性构建原始样本集,采用bootstrap技术对原始样本集重取样,利用粗糙集约简算法约简重取样出的样本特征,然后用约简后的样本训练成员回归机、建立预测模型,将各成员回归机的预测结果采用均值法融合输出,其输出即为预测的色差平面上待插值点的色差,最后根据预测的色差值计算出丢失的像素值。仿真实验结果表明:所提去马赛克方法获得较优的客观指标彩色图像峰值信噪比(rCPSNR)和S-CIELAB的色差 ,较好的保护图像的细节信息,达到较满意的视觉效果。

Abstract: A support vector machine (SVM) ensemble based demosaicing algorithm was proposed, which can reduce edge artifacts and false color artifacts effectively, and improve the image edge feature. Firstly, original sample set was constructed by applying inter-pixel correlation on green-red or green-blue color difference plane. The color difference plane was constructed by applying inter-channel correlation of the R, G, B channels and the constant color difference principle. Secondly, the training samples were selected by using the conventional bagging algorithm was used to re-sample the original sample set, and use the rough set reduction algorithm to dynamically reduce sample characteristics of the re-sampled samples. Thirdly, the individual support vector regression (SVR) was trained with the reduced training samples, and forecast the unknown color difference between two color channels (green-red or green-blue was forecast). Finally, the missing color value at each interpolated pixel was calculated using the forecasted color difference. Simulation results show that the proposed approach produces visually pleasing full-color result images and obtains higher rCPSNR and smaller colour difference of S-CIELAB than other conventional demosaicing algorithms.

有色金属在线官网  |   会议  |   在线投稿  |   购买纸书  |   科技图书馆

中南大学出版社 技术支持 版权声明   电话:0731-88830515 88830516   传真:0731-88710482   Email:administrator@cnnmol.com

互联网出版许可证:(署)网出证(京)字第342号   京ICP备17050991号-6      京公网安备11010802042557号