Abstract:

Gradient is an important feature of image. Local Phase Quantization（LPQ） model for texture classification based on multi-directional gradient is proposed. The texture local phase quantization model based on multi-directional gradient first extracts the gradient features from different directions of image, and then encodes the gradient features in each direction using the local phase quantization method. The gradient features in each direction coded by LPQ are connected into a matching vector. In order to make full use of the gradient information of the image, the local phase quantization method with block mode is also discussed. Experiments on two texture databases and one palmprint database fully show that local phase quantization coding for gradient information in several directions of image is an effective texture feature extraction algorithm.

Key words: multi-directional gradient of image, block local phase quantization, texture feature extraction

摘要：

梯度是图像的一种的特征，而同时考虑不同方向上的梯度信息是一种更加有效利用梯度的方式，因此提出多方向梯度的纹理局部相位量化模式算法。多方向梯度的纹理局部相位量化模式首先从不同方向提取图像的梯度特征，然后对每个方向上的梯度特征采用局部相位量化方法进行编码，各方向梯度采用相位量化编码后的特征连接成一个匹配特征向量。为了充分利用图像的梯度信息，还探讨了块模式的局部相位量化方法。两个纹理数据库和一个掌纹数据库上的实验充分表明，对图像各方向上的梯度信息进行局部相位量化编码是一种有效的纹理特征提取算法。

关键词: 图像多方向梯度, 块局部相位量化, 纹理特征提取