Image quilting纹理传输算法的改进研究

An Improvement to the Image Quilting Texture Transfer Algorithm

纹理合成与纹理传输是目前计算机图形图像领域比较热门的研究方向.目前针对纹理传输的研究大部分都集中在纹理块与目标图像对应像素颜色信息的匹配度的改进上,目标图像整体的结构信息鲜有提及.以Efros等人提出的image quilting纹理合成算法为基础,提出一种新的纹理传输改进算法,新算法主要对纹理传输过程中的误差匹配公式作了改进,引入了与目标图像的边缘梯度信息有关的项,同时还采用了亮度重映射方法对纹理块进行匹配预处理以解决因纹理图像和目标图像亮度差过大而带来的纹理块搜索困难问题.广泛的实验表明改进后的纹理传输算法在时间上和image quilting算法相差无几,但传输后的图像整体轮廓清晰、亮度明显,因此传输效果更合理、更加符合人的视觉感受.

Texture synthesis and texture transfer are popular research directions in the field of computer graphics and imaging. Present researches on texture transfer mostly aim at the improvement of the match strategy of the color information between the corresponding pixels of the texture block and the target image,and the structural information of the target image is seldom involved. In this paper,we proposed a newtexture transfer algorithm based on the original image quilting texture transfer algorithm by Efros et. al. The proposed algorithm improves the error matching formula by adding a term relating to the edge gradient information of the target image. Luminance remapping method is also introduced as a preprocessing stage to diminish the matching texture block searching ghost caused by too large luminance difference between the texture and the target images. Extensive experimental results showthat the proposed algorithm can produce outline clearer transferred image with augmented brightness but with roughly the same time consumption compared with the original texture algorithm by Efros et. al. Therefore,the resulted transfer effect is more reasonable and in more consistent with human visual perception.