基于DCT域纹理特征的多聚焦图像融合  

Multi-focus Image Fusion Based on Texture Features in DCT Domain

作　　者：唐振华[1,2] 梁聪[1] 区骋 黄旭方[1] 覃团发[1,2] 

机构地区：[1]广西大学计算机与电子信息学院,南宁530004 [2]广西大学广西高校多媒体通信与信息处理重点实验室,南宁530004 [3]中国铁塔广西区分公司,南宁530007

出　　处：《应用科学学报》2015年第6期628-636,共9页Journal of Applied Sciences

基　　金：国家自然科学基金(No.61461006;No.61261023);广西省自然科学基金(No.2013GXNSFBA019271)资助

摘　　要：现有的基于离散余弦变换(discrete cosine transform,DCT)的多聚焦图像融合算法容易使融合图像出现块效应和伪影,为此提出一种基于DCT域纹理特征的图像融合算法.该算法以8×8 DCT块中反映能量方向性的纹理区域作为图像融合单位,根据纹理区域的频谱相似度,选择能量较大的区域或以区域加权叠加的方式获得融合区域.实验结果表明,与现有基于DCT域的多聚焦图像融合算法相比,该算法获得的融合图像主观质量较好,能有效避免明显的块效应与伪影.Exiting multi-focus image fusion methods based on discrete cosine transform （DCT） domain cause blocking effects and artifacts in the fused images. To address this problem, an image fusion algorithm concentrated on texture features in the DCT domain is proposed. The proposed method separates each 8×8 DCT block into several texture regions having various directivity of energy distribution in the source images. Similarity of the frequency spectra for the two texture regions are calculated. The part of the fused image is obtained by selecting the one with more energy, or using the weighted result. Experimental results show that the proposed algorithm can effectively reduce blocking effects and artifacts. Compared with exiting approaches in DCT domain, the proposed algorithm can provide fused images with better subjective quality.

关 键 词：多聚焦 图像融合 DCT域 纹理特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]