基于图切割的拉普拉斯金字塔图像融合算法

An image fusion algorithm of Laplacian pyramid based on graph cuting

摘要:

针对在图像拼接中普通的拉普拉斯金字塔融合算法 容易丢失细节的缺点,以及由于运动物体造成的 融合鬼影现象,本文提出了一种基于图切割的拉普拉斯金字塔融合算法。首先,引入图切割 技术,寻找最 优缝合线,确定一种自适应融合区,以消除运动物体造成的融合鬼影；其次,利用源图像完 整细节对重构 误差进行补偿,提出一种基于包含水平方向在内多个方向的加权融合方法,将源图像 和拉普拉斯金 字塔融合图像按照这种融合规则进行融合。实验结果表明,与经典拉普拉斯融合方法对比, 在客观指标上,本文方法的图像均值平均提高了0.326,标准差(SD)平均提高了1.109,信 息熵平均提高了0.041,图像清晰度平均提高了0.289；在主观效果上,本文方法无明显拼接痕迹和融合鬼 影,较好保留了图像细节,提高了融合质量,全景图拼接更加真实,改善了整体的视觉效 果。

Abstract:

Due to the classical Laplacian pyramid image fusion algorithms are in loss of image details and moving objects resulting in fusion ghosting in image stiching,an image fusion al gorithm of Laplacian pyramid based on graph cuting is proposed in this paper.Firstly,an optimal seam is found via g raph cuting introduced.Then an adaptive transition region is proposed to remove the fusion ghosting.Secondly, the reconstruction error is compensated by the details of the original image.The weighted fusion algorithm i s proposed with consideration of multi- direction,including horizontal direction.The source images and the fused images of Laplacian pyramid are merged together according to the fusion rules.Compared with classical Laplacian fusion method,experimental results show that the method proposed in this paper performs best in terms of subjective quality and objective indicators. Objectively,it improves image average by an average of 0.326,standard deviatio n by an average of 1.109, information entropy by an average of 0.041and image definition by an average of 0.289.Subjectively,it retains the image details and then improves the quality of fusion without obvious stitch ing artifacts and fusion ghosting.The panorama stitching is more true and the ov erall visual quality is improved.