目的 现实中的纹理往往具有类型多样、形态多变、结构复杂等特点，直接影响到纹理图像分割的准确性。传统的无监督纹理图像分割算法都具有一定的局限性，不能很好地提取稳定的纹理特征。方法 本文提出Gabor滤波器和扩展的Local Ternary Pattern （LTP）算子来描述纹理的差异性特征。具体的，利用Gabor滤波器提取相同或相似纹理模式的纹理特征，这时，提出扩展的LTP算子提取纹理的差异性特征，并将这些特征融入到水平集框架中，进行纹理图像的分割。本文提出的方法优势体现在：一方面，扩展的LTP算子能够有效的提取局部区域的纹理差异性特征；另一方面， Gabor滤波器和扩展的LTP算子具有互补性。结果 通过实验证明，对于纹理方向及尺度变化较大的图像、复杂背景下的纹理图像、弱纹理模式的图像，本文所提出的方法整体分割结果明显优于传统的Gabor滤波器、结构张量、拓展结构张量、局部相似度因子等纹理分割方法所得到的结果。同时，将本文方法与基于LTP的方法进行了对比，本文方法的分割结果依然更优。更直观的是，本文给出具体的分割准确度的量化结果进行对比，本文方法与各种无监督的纹理分割方法进行了对比，在典型的纹理图像上，本文方法准确度达到了97%以上，高于其它方法的分割准确度。结论 本文提出了一种无监督的基于多特征的纹理图像分割：基于Gabor滤波器和扩展的LTP算子，能够较好的提取相似纹理模式的特征和纹理的差异性特征，且将这些纹理特征很好地融合到水平集框架中。实验显示，本文提出的方法对于真实世界复杂纹理图像能够得到良好的分割效果。
Unsupervised segmenting texture images based on Gabor filters and extended LTP operator
Ma Rui,Zhou Li(Hefei University of Technology)
Objective In reality, the texture is often characterized by various unregular types, varied shapes and complex structures, which directly weaken the accuracy of texture image segmentation. The semantic segmentation methods based on deep learning need Benchmark training sets, and it is difficult to construct the training sets composed by the complex and diverse texture images. Therefore, it is needed to utilize the unsupervised image segmentation methods to solve the problem of texture segmentation. However, the traditional unsupervised texture image segmentation algorithms have some shortcomings and cannot used to extract the stable texture features well. Method Based on the idea that the Gabor operator can extract the texture diversity features and the LTP operators imply the threshold differences, we propose to combine Gabor filters with extended LTP operators to describe texture diversity features in this paper. Specially, we use Gabor filter to extract the texture features of the same or similar texture patterns. Then, we extract the texture difference features by proposing. Compared with the traditional LTP, the main advantages of the extended LTP operator are embodied in two aspects. On the one hand, according to the size features of the segmented image, the extended size make the LTP operator effective in image segmentation. On the other hand, the weights are given to each position of the extended LTP operator. Here, the exponential weight differences are given according to the distances between each position and the central point. Finally, these exacted features are integrated into the level set frame to segment the texture image. The advantages of the proposed method in this paper are described in the following: First, the extended LTP operator can effectively extract the texture difference features of local regions. Second, the Gabor filter and the extended LTP operator are complementary. The main contributions of the proposed method in this paper are elaborated as follows:
1. By improving the traditional LTP operator, an extended LTP operator is proposed to extract the texture difference features of pixels in complex images;
2. The Gabor filter and the extended LTP operator are complementary. The extended LTP operator and the Gabor filter are combined and incorporated into the level set method. The extended LTP operator extracts the texture difference information of complex images, while the Gabor method detects the similar information such as similar frequency, size and direction of images. The two operators have obvious complementary characteristics. Therefore, the extended LTP operator and the Gabor filter are combined to extract the texture features of complex images in a complementary way. The two operators are integrated into the level set method to effectively solve the segmentation problem of the complex images.
Result In the section of experimental results, we compare the proposed method with the classical unsupervised texture segmentation methods, including the method based on Gabor filter, the method based on structure tensor, the method based on extended structure tensor and the local similarity factor. By segmenting all kinds of texture images such as the images with varied texture directions and sizes, the images with the complex background and the images with the weak texture features, we testify that the proposed method achieve the better segmentation results than that of the traditional Gabor filter, the structure tensor, expand the structure tensor, the local similarity factor. Meanwhile, by comparing the proposed method with the LTP-based method in this paper, we also testify that the segmentation results of the proposed method are still better than that of the LTP-based method. In the section of experimental results, the segmentation results of some commonly used level set methods (including: Gabor filter, structure tensor, extended structure tensor (LCV), robust local similarity operator (RLSF)) are presented, and compared with the segmentation results of the proposed method. In detail, the advantages of the proposed method for segmentation of three types of texture images are shown in FIG. 8, FIG. 9 and FIG. 10 respectively. More intuitively, we give specific quantitative results of segmentation accuracy for comparison between the proposed method and the various unsupervised texture segmentation methods in this paper. On some typical texture images, the accuracy of the proposed method is more than 97%, which is higher than that of the other methods. Conclusion We proposes an unsupervised multi-feature-based texture image segmentation algorithm in this paper. The proposed method can be utilized to extract the features of the similar texture patterns and the texture differences by combining the Gabor filter and extended LTP operator. And then we integrate these texture features into the level set framework to segment the texture images accurately. Lots of experiments show that the proposed method can achieve desirable segmentation results for complex texture images in real world. We analyze the advantages and disadvantages of the proposed unsupervised segmentation method and the methods based on deep learning in the section of conclusion. Compared with the segmentation methods based on deep learning, the proposed method is an unsupervised method, which does not need prior information or training information. However, the segmentation methods based on deep learning relies on training information heavily. It is a difficult task to obtain the training information of the complex texture images. At the same time, the future research ideas are elaborated. Specially, considering some structural relevance of texture images, our future work will mainly focus on the extraction and analysis of texture structures in order to obtain better segmentation effects.