综合纹理和形状特征的外观专利图像检索算法研究

Comprehensive texture and shape features algorithm research for appearance pattern image retrieval

针对外观设计专利图像背景多样性、复杂性以及形状特征突出等特点,提出了一种综合纹理和形状的检索算法。该方法首先采用基于物体内部结构纹理特征描述纹理,并用欧式距离取得其相似值,然后再用改进的加权欧式距离取得不变矩形状特征向量的相似值,经特征融合得到最终相似距离值。实验结果表明,该算法优于现有的其他算法,针对外观专利图像的检索,具有更高的查全率和查准率。

Considering diversity, complexity and prominent shape feature of the appearance design pattern images, a retrieval algorithm of comprehensive texture and shape features is proposed in this paper. In this method, the texture charaeteristies based Oil object interior structure to describe the texture is firstly generated, and obtains image similar value by using Euclidean distance. Then it uses the improved weighted Euclidean distance formula to get the invariant moments shape characteristic vector of the simi- lar values. Ultimately, it gain similar distance through the characteristic fusion. The experimental results show that the method is better than other existing algorithm. Aiming at the appearance patent image retrieval, it has higher recall ratio and precision ratio.