目的 复杂纹理的图像分割一直是图像分割的难题，现有的一些纹理图像分割的方法提取图像的确定方向的灰度变化特征或者提取图像的局部灰度相似性特征，然而，自然纹理中普遍存在局部纹理的形态相似性和方向不确定性的现象，导致现有的纹理图像分割方法不能准确的分割纹理图像。方法 本文提出局部连接算子和局部差异算子来描述局部纹理的形态相似性和局部纹理的差异度；具体的，一方面，通过设定一定的阈值，将局部区域的灰度差异分为两类，分析两类差异的分布特征，从而提取图像的形态特性并得到局部连接度算子；另一方面，设置一种无方向性的灰度差异分析算子，提取图像局部的灰度差异值从而得到局部差异度算子；两个算子结合以更好地提取纹理图像的局部特征，然后通过融合局部相似度特征，局部差异度特征和灰度信息，构造水平集能量泛函,进而通过最小化能量泛函实现纹理图像分割。 结果 通过实验发现：相比基于Gabor变换、结构张量、局部相似度因子的纹理分割方法，提出的局部算子能够更好的区分自然图像的不同的纹理区域，这时，通过实验分析发现，提出的方法对论文中实验图像的的平均分割准确率高达97%，远高于其他方法。因此，提出的模型对于自然纹理图像具有更好的分割效果。结论 本文提出了两种新颖的纹理特征局部描述子：局部连接度算子和局部差异度算子，两种算子能够有效的提取纹理特征，且有一定的互补性。实验表明，提出的方法对于复杂自然纹理图像具有良好的分割效果。
Objective It is a difficult problem to segment images with complex texture features. Many former methods, such as Gabor filter and structure tensor, have been proposed to segment texture images. However, some drawbacks existed in these methods. For example, the method based on Gabor filter need to set multiple scales and directions and it is difficult to select the desirable scale and direction parameters. The method based on structure tensor set two fixed direction to analyze the intensity variations. However, many texture images include unregular direction feature, and so on. Besides, it is not enough to only utilize some intensity variation information. In fact, the morphology similarity and direction uncertainty are often existed in nature texture images. The existed methods cannot be utilized to segment nature texture images accurately. Method Considering the morphology similarity and direction uncertainty of local texture pattern, we propose the local connection and difference operators to extract the texture features of images. On one hand, by setting threshold for each local region, the local connection operation can be used to analyze the intensity distribution and texture morphology features. On the other hand, considering the intensity variation trait of texture image, the local difference operation can be utilized to extract the local intensity variation. The local connection operation and local difference operation are complementary. Then, by combining local similarity feature, local difference feature with intensity information, the level set energy functional is constructed and further minimized to obtain the final segmentation results. The main advantages of the proposed method can be summarized as the following two points: First, the morphology feature is proposed to analyze the local intensity distribution feature of texture images. It should be noticed that the intensity distributions of texture images are uncertain for each object region. Thus, it is a meaning work to extract the morphology feature of local region. Second, the proposed two operators are complementary. It is difficult to utilize only a feature to segment complicated texture images. The morphology feature and intensity difference feature can jointly to segment images so that the proposed method can be robust for different texture images. Result We verify the effectiveness of the two operators by comparing with other operators such as Gabor, structure tensor, extended structure tensor and local similarity factor. Specially, we firstly analyze the extraction effects of morphology feature by comparing the method based on Gabor filter, these methods based on structure tensor and extended structure tensor, the method based on local similarity factor with the proposed local connection operator. Obviously, the extracted feature of local connection operator can be used to discriminate the object and background regions efficiently. Then, the local difference operator is also used to compare with the traditional texture feature extraction methods. It is obvious that the local different operator can extracts the local intensity variation accurately. In next, we make three comparison experiments (Fig.9-Fig.11) to testify that the proposed method can achieve better segmentation effect than these methods based on Gabor, structure tensor, extended structure tensor and local similarity factor. Finally, we testify the robustness of the proposed method for different initial contours and images. Besides, the segmentation accuracy of our proposed method can higher than 97%, which is much higher than other methods. That is, it is better to discriminate the different texture patterns of images by using the proposed operators. Conclusion We propose two complementary operators: the local connection operator and local difference operator to extract texture features effectively. Meanwhile, the extracted two features are complementary and can be jointly utilized to segment texture images. Finally, we testified that the proposed method can obtain better segmentation results for natural texture images.