Current Issue Cover


摘 要
目的 纹理滤波是计算机视觉领域的一个基础应用工具,其目标是抑制图像中不必要的纹理细节和保持图像的主要结构。目前已有的纹理滤波方法多存在强梯度纹理无法被抑制或结构丢失的问题,为此提出一种结合纹理梯度抑制与L0梯度最小化的纹理滤波算法。方法 首先,提出一种能够区分结构\纹理像素的方向性区间梯度算子,其中采取了局部对比度拉伸和尺度自适应策略,提升了弱梯度结构像素的识别能力。随后,利用区间梯度幅值对原始图像梯度进行抑制,并用抑制后的图像梯度进行图像重建,获得纹理像素梯度小于结构像素梯度的纹理抑制图像。最后,考虑到纹理梯度抑制时会对结构像素的梯度产生一定的衰减作用,本文采用具有梯度提升作用的L0梯度最小化方法对纹理抑制图像进行滤波,得到纹理抑制结构保持的纹理滤波图像。结果 通过测试马赛克和自然风景等不同类型的图片,并与L0梯度最小化,RGF,RTV,IG,COF等方法相比较,本文算法能够在抑制强梯度纹理的情况下对图像的主要结构得以保持,并且具有良好的普适性和鲁棒性。同时本文将纹理滤波应用于图像的边缘检测和细节增强,取得了不错的效果提升。结论 本文算法在兼顾强梯度纹理的抑制和结构的保持方面已超越已有的方法,对于图像的目标识别、图像融合、边缘检测等易受强梯度纹理干扰的技术领域,具有较大的应用潜力。
Texture filtering by using texture gradient suppression and L0 gradient minimization

Shao Huan,Liu Chunxiao(School of Computer Science Information Engineering,Zhejiang Gongshang University)

Objective Texture is a repetitive pattern with high pixel values, many natural images and works of art include textures such as cross-stitch and mosaic. In many cases, people"s visual system will ignore the texture pattern and pay more attention to the main structure of the image. Texture filtering is a basic tool in the field of computer graphics and image processing, whose goal is to suppress unnecessary texture details and maintain the salient structure in the image. In recent years, various texture filtering methods has been proposed, which are mainly divided into global and local-based filtering methods. Most of the existing texture filtering methods tends to deal with the small gradient texture images, however, it is difficult to handle the strong gradient texture and lose part of structure. In order to solve this problem, we propose a texture filtering method by using texture gradient suppression and L0 gradient minimization, which can achieve to suppress texture and maintain structure.Method The main idea of this algorithm is to obtain the input image with strong gradient texture suppression, and then obtain the smooth filtering results by the traditional texture filtering method, in which the traditional texture filtering method uses L0 gradient minimization. Our method contains three steps to achieve the goal of image filtering. First of all, we improve the interval gradient operator, which has the ability to distinguish texture and structure pixels. We propose a directional interval gradient operator to increase the gradient amplitude by finding the main direction of the structure. Since the pixel gradient value of the weak structure area will be much smaller than the gradient value of the strong gradient texture, we use the local contrast stretching strategy when calculating the direction interval gradient to improve the recognition ability of the weak gradient structure. Directional interval gradient affects the effect of texture suppression, so the selection of computational scale is particularly important. In this paper, a scale adaptive strategy is proposed, which automatically selects the optimal scale for calculating interval gradient. Secondly, we want to get an input image with texture gradient suppression. In the first step, we get the directional interval gradient value that the structured pixel is larger than the texture pixel. And then the normalized directional interval gradient amplitude is used as the basis to attenuate the gradient of the original image. Image reconstruction is performed after image gradient suppression to obtain a texture-suppressed image having a gray pixel gradient smaller than the structural pixel gradient. In the image reconstruction step, we transform the reconstruction problem into the function optimization problem, and transform the fast Fourier transform to the frequency domain to solve the problem. In the end, considering that the texture gradient suppression operation will cause some attenuation of the gradient of the structural pixels and thus cause the loss of the structure, we use the L0 gradient minimization algorithm with gradient lifting effect to filter the reconstructed image to remove texture while preserving structure.Result To prove the validity of the proposed method, we tested different types of pictures, including mosaics, nature, grasslands, etc. All the experiments are run on the windows platform, and the algorithm is implemented in MATLAB. In this paper, there are three main parameters to be set, which are the scale of the calculation of the directional interval gradient, the gradient weight η in the image reconstruction step and the smooth parameter λ in the L0 gradient minimization. η controls the suppression degree of the strong gradient texture in the reconstructed image. With the increase of η, the texture is suppressed better. In order to evaluate the performance of our method, we compare it against other texture filtering models, including L0 gradient minimization method, rolling guidance filtering method, interval gradient method, co-occurrence filter method, and relative total variation method. All methods use the code provided by the author and debug the optimal parameters to get the filter result. There is no reasonable objective evaluation index in the field of texture filtering. As a consequence, the subjective evaluation of human eyes is used to compare the effects of different methods. The experimental results can be summarized as follows. In the mosaic image with strong gradient texture information and intractable tiny structures, this algorithm surpasses the effects of other algorithms in strong gradient texture suppression, and the small gradient structure is also maintained. When processing natural images, our algorithm demonstrates superior image smoothing results in filtering out the vary scale textures and preserving the small gradient structures. At the same time, the texture filtering is applied to image edge detection and detail enhancement, which achieves a good effect.Conclusion Aiming at the trade-off between strong gradient texture suppression and structure preservation in current texture filtering methods, this paper proposes a texture filtering algorithm that combines texture gradient suppression and L0 gradient minimization. Its main purpose is to first suppress the texture gradient of the input image, and then the texture and structure pixels have different texture filtering operations. The experiments demonstrate that our algorithm can maintain the main structure of the image and make the gradient smoother. In the field of image recognition, image fusion, and edge detection that are susceptible to strong gradient textures, texture filtering has great potential for application.