基于分形维数向量的图像类推技术

Image analogies technology based on vector of fractal dimension

为了给出图像类推算法的适用范围,提出了一种新的基于分形维数向量的图像的相似性度度量算子。对图像类推算法进行了认知和分析,指出图像类推在本质上是一个多分辨的纹理合成问题,它对图像的结构和素材都存在着较高的要求。通过实验结果表明,提出的基于分形维数向量的图像相似度度量算子可以较好的表现图像的结构和素材这两个特征。并通过实例验证了提出的图像相似度算子较好的解决了原有Image Analogies方法适用的模糊性问题,从某种意义上揭示了风格化学习和风格化继承的特性。

To give applicable range of image analogies algorithm, based on the fractal dimension vector, a new operator used to measure the image similarity is proposed. First image analogies are analyzed and classified, a viewpoint which image analogies is essentially more than resolution texture synthesis is put forward, also it have the high request to the structure and color of images. A new operator which can highly measured the structure and color ofimages is proofed through the experiment. Through this new operator the applicable range of image analogies algorithm is given, also the characteristic of the style learne and the style inherits is promulgated by some way.