目的 本文基于现有的研究提出一种细节感知的纹理去除算法，在去除图像纹理时，能够很好的保持图像的结构信息，尤其是诸如细长结构和边角信息等在其他方法中容易被模糊化的特殊细节。方法 首先，本文将提出一种能够识别细长结构的结构检测方法，对细长结构检测并增强。其次，为了估计每个像素点的最优滤波核尺度，改进原有的相对总变差模型，多方向寻找最小相对总变差，使它能够更好的区分纹理和边界，并且将边角信息从纹理中区分出来。然后，将检测出来的细长结构归一到改进的相对总变差的度量尺度上，估计滤波核尺度，生成引导滤波图像。这样就能够在平坦或有纹理的区域运用大尺度的滤波核，并在结构边缘和边角附近减小滤波核。最后，通过联合双边滤波器得到纹理去除后的图像。结果 实验测试了马赛克图像和艺术画作，对比了相对总变差和尺度敏感的结构保护滤波等方法，本文方法在去除纹理同时保留了细长结构和边角细节，并且具有良好的普适性和鲁棒性。处理一张含十万像素的图像，本文算法通过一次迭代计算就能够去除大量纹理且效果优于已有的方法，本算法的计算时间为3.37秒，其它算法为0.07至3.29秒。 结论 本文设计的纹理滤波器不仅在保持诸如细长结构方面的性能更好，而且使纹理去除后图像的边角细节处更尖锐，为图像的后续处理提供了一种强有力的图像预处理方式。
Detail-aware texture filtering algorithm
Xiao Yi,Zhu Xianyi,He Yangti,Zheng Yan(College of Computer Science and Electronic Engineering,Hunan University,ChangSha;Hunan university)
Objective This paper proposes a detail-aware texture removal algorithm based on existing studies. When removing the image textures, the proposed method can maintain the fine structural information of the image, especially the special details that are easily obscured in other methods such as the slender structure and corner information. With the continuous development of computer technology, the application of image processing technology has become more and more widespread in areas such as pattern recognition, security monitoring, smart driving, and computer photography. However, the image quality collected directly from the image acquisition card is not particularly satisfactory. Therefore, image pre-processing is necessary. In the pre-processing of the image, texture filtering is an important step. In the texture image, the image edge is the main component of the image structure. The traditional image filter processing techniques such as median filtering and Gaussian filtering can filter noise to a certain extent, but the structure context will also be filtered. Therefore, this paper investigates how to remove texture and maintain slender structure context simultaneously. Method The main idea of this algorithm is to compute the optimal filter scale by leveraging a novel slender structure recognition technique and an improved structure angle relative total variation based on multiple direction, then obtains the filtering result through guided filter. Specifically, the proposed method consists of four steps. First of all, according to the deficiency of the existing algorithms this paper proposes a method that can identify and enhance slender structures to avoid smoothing them in the subsequent texture filtering process. Secondly, in order to estimate the optimal filtering kernel scale of each pixel, the original relative total variation model is improved by searching the minimum relative total variation in multiple directions so that it is able to distinguish textures and boundaries more effectively and corner information is better distinguished from texture. Then, the detected elongated structure is normalized to the improved metric of relative total variation, and the filtered kernel scale is estimated to generate a guided filtered image. So that large-scale filtering kernels are used in flat or textured regions, and the filtering kernels are reduced near the edges and corners of the structure. Finally, the texture-removed image is obtained by combining the joint bilateral filters. Result We evaluated our method on different types of pictures, including mosaics and paintings, etc. The experiments are conducted on a Windows 8.1 operation system, and the proposed method is implemented in MATLAB language. Since there do not exist any reasonable quantitative objective evaluation metrics so far in the research field of texture filtering, the subjective evaluation by human eyes is commonly used. In the experiments, we compared the five existing methods of texture filtering, including bilateral texture filtering, rotation guided filtering, relative total variation, scale-sensitive structural protection filtering and interval gradient operator. Compared with these five methods, the proposed algorithm needs a slightly longer computing time. More specifically, for the image with 394×304 pixels, the proposed method takes 3.37s, while these methods takes 2.23s, 0.07s, 0.23s, 1.01s and 3.29s respectively. Fortunately, our method outperforms these methods in terms of texture removal while maintaining the slender structures and corner details. In the experimental parts, we also analyzed the iteration number and the parameter value standard deviation σ of the proposed algorithm. The results of the comparative experiments demonstrate that with one iteration used to remove the texture, the result of our algorithm is better than the two methods: relative total variation and interval gradient operator. For standard deviation σ, larger σ is chosen when the optimal filter scale is larger, and smaller σ is equipped when the optimal filter scale is smaller. Conclusion The texture filter designed in this paper performs better in maintaining features such as elongate structures and sharp corners in the image after texture removal, which provides a powerful image pre-processing method for image subsequent processing, including image detail enhancement, edge detection, image abstraction, and image segmentation. For the problem of sharp reduction of the filter kernel scale encountered in the experiment, this paper makes a reasonable explanation and gives the solution. The shortcoming of the proposed algorithm in this paper is the long computing time. In general, although sacrificing a very small computing time efficiency, it is worthwhile to get better results.