基于LBP和KNN的视频纹理识别算法

为实现视频纹理的有效识别, 提出一种基于 LBP(Local Binary Patterns)和 KNN(k-Nearest Neighbor)的视频纹理识别算法。 该算法将视频纹理视为一个图像纹理集合, 通过多个图像纹理集合的方式表示。 由于可计算任意两幅纹理图像的相似度, 对于两个视频纹理, 可以计算两个图像纹理集合中所有元素之间的相似度, 将这些相似度中的最小值作为这两个视频纹理的相似度, 若要实现视频纹理的识别, 则可通过 KNN 算法实现分类与匹配。 通过在 DynTex 动态纹理数据库中的相关实验, 证明了该算法的有效性。

Video Texture Recognition Based on LBP and KNN

To achieve the effective recognition of the video texture, we propose a video texture recognition algorithm based on LBP (Local Binary Patterns) and KNN (k-Nearest Neighbor), which can be regarded as an image set, and a video texture can be represented by an image set. For the two video textures, the similarity between every elements of the two image sets is calculated, and the minimum value of the similarity is traited as the similarity of the two video textures. Then the classification and matching of test video texture can be realized by KNN algorithm. The validity of the proposed algorithm is verified by the experimental verification of the DynTex dynamic texture database.