基于特征提取的纹理合成算法

Texture Synthesis Method Based on Feature Extraction

对于包含明显特征对象的纹理,现有的纹理合成算法会破坏样图中特征对象的完整性,使得合成的纹理效果不理想。针对该类型的纹理,本文提出了一种基于特征提取的纹理合成算法,该算法将纹理样图进行分层处理,一层是纹理上具有明显边界和完整形状的特征对象,一层是除去特征对象的背景。我们先将纹理上的特征对象提取出来,然后用剩下的纹理合成背景,最后将提取的对象按照样本中的分布规律撒播在背景上来生成纹理。实验表明,该算法能够较好地保证样本纹理中特征对象的完整性,合成效果比较理想。

As for the texture with obvious characteristics, the texture produced by the present texture synthesis algorism, mostly, will break the wholeness of characteristic objects in the sample, making the synthesizing effect far from our expectation. As far as this type of texture is concerned, this paper presentes a texture synthesis method based on feature extraction. This algorism makes a processing in layers to texture samples, one layer being complete-shaped characteristic object with a obvious frame and another getting rid of the background of the characteristic object. We firstly extract the characteristic object from the texture, then synthesize to the background with the rest texture and finally broadcast the extracted objects to the background to produce texture according to the distributing regulations in the sample. As the experiment proves, this algorism can assure the wholeness of characteristic object in the sample texture very well, and the synthesis effect is relatively good.