A new efficient method to characterize dynamic textures based on a two-phase texture and dynamism analysis
A new efficient method to characterize dynamic textures based on a two-phase texture and dynamism analysis
复制标题
DOI：
10.1016/j.patrec.2014.04.009
复制
发表时间：
2014-08
期刊：
影响因子：
--
通讯作者：
Mina Koleini;M. Ahmadzadeh;S. Sadri
中科院分区：
文献类型：
--
作者：
Mina Koleini;M. Ahmadzadeh;S. Sadri
动态纹理（DT）是纹理到时域的扩展。近年来，DT的描述和分类引起了广泛关注。本文提出了一种新的 DT 分类和合成方法。该方法基于两相纹理和动态分析。首先，提出了一个数学模型来模拟 DT 的动态性。然后通过两步过程来描述 DT：描述 DT 的纹理和动态。双树复小波变换 (DTCWT) 应用于 DT 动态的纹理帧和模型。这使得我们的算法对于光照和位移变化足够鲁棒。从纹理帧和 DT 动态模型获得的复小波系数的平均值和标准差被连接并用于形成 DT 特征向量。通过使用特征向量的傅里叶变换，还提供了旋转不变性。然后使用由这些特征向量组成的视觉单词词典来描述每个 DT。这两个阶段共同涵盖了各种各样的 DT 分类问题，包括类别在外观和运动方面都不同的情况，以及所有类别的外观相似但只有运动具有判别性的情况，反之亦然。与早期的方法相比，我们的方法有很多优点，例如为两个测试数据库提供高速和更好的性能。
Dynamic texture (DT) is an extension of texture to the temporal domain. Recently, description and classification of DTs have attracted much attention. In this article, a new method for classifying and synthesizing DTs is proposed. This method is based on a two-phase texture and dynamism analysis. At first, a mathematical model is proposed to model the dynamism of a DT. Then a DT is described in a two-step procedure: describing the texture and the dynamism of a DT. Dual Tree Complex Wavelet Transform (DTCWT) is applied on the textured frames and models of the dynamism of the DT. This makes our algorithm robust enough to illumination and shift variations. The mean and standard deviation of the complex wavelet coefficients, as obtained from textured frames and models of the dynamism of the DT, are concatenated and used to form the DT feature vector. By using Fourier Transform of the feature vector, rotation invariance is also provided. A dictionary of visual words made from these feature vectors is then used to describe each DT. Together, the two phases cover a large variety of DT classification problems, including the cases where classes are different in both appearance and motion and those where appearance is similar for all classes and only motion is discriminant or vice versa. Our approach has many advantages compared with the earlier ones, such as providing high speed and a better performance for two test databases.