基于自适应分数阶阶次的图像增强和图像匹配

发布时间：2018-04-13 14:30

本文选题：图像增强 + 图像匹配　；参考：《南昌航空大学》2017年硕士论文

【摘要】：图像增强和图像匹配是图像处理和计算机视觉领域的两个基础性的重要研究内容。传统的图像增强和图像匹配方法大多数是基于整数阶微积分,对图像中存在弱导数性质的弱边缘和弱纹理效果不理想。分数阶微分具有在增强图像的同时可以更好地保留图像中的弱边缘和弱纹理细节信息的优良特性,但已有的基于分数阶微分的图像增强方法,需要通过人工寻找最佳阶次,缺乏微分阶次自适应性;在图像匹配中,将SIFT(Scale Invariant Feature Transform)匹配算法与分数阶微分相结合,在模糊图像和弱纹理图像中能提取到更多的特征点,从而提高了匹配的精度,但是最佳微分阶次的选择仍然需要人工调整,费时费力。因此,本文针对这两个问题展开研究,具体工作如下:1.通过分析分数阶微分对信号的作用,构造了在分数阶微分图像增强中的微分阶次自适应模型,该模型以反正切函数为原型,以图像的梯度信息、局部信息熵、亮度和对比度为自变量,建立了分数阶微分阶次与图像局部信息之间的关系,从而可以根据图像的局部特征信息自动计算图像中各个像素点的最佳阶次,并将该模型应用到分数阶微分Tiansi算子的图像增强中。为了验证本文模型的有效性,选用标准图像库中的多幅纹理图像进行实验,对实验结果进行了定性和定量分析,并与二阶微分Laplacian算子,Tiansi算子进行比较。理论分析和实验结果均可表明本文建立模型的有效性,对灰度图像可以得到持续变化的增强效果,接近于最佳分数阶微分阶次的增强实验效果,符合人类在视觉上的感受。2.提出了一种自适应分数阶的SIFT算法,用于图像匹配。算法在Riemann-Liouvill(R-L)分数阶微分的基础上,设计了一种新的分数阶微分掩膜,并将其融入到SIFT算法中,提取到更多精确有效的关键点,从而提高了SIFT算法的匹配精度;然后根据图像的局部特征,构造了分数阶微分阶次自适应数学模型。该模型以反正切函数为原型,以图像的梯度信息、局部信息熵为自变量,建立了分数阶微分阶次与图像局部特征信息之间的关系,从而可以根据图像的局部特征信息自动计算图像中各个像素点的最佳阶次;为了验证本文算法和模型的有效性,选用标准库中的图像和真实图像进行实验,与原始SIFT算法和基于分数阶微分的SIFT算法进行比较;并对算法的效率和最佳微分阶次进行分析,理论分析和实验结果均表明本文算法的有效性。
[Abstract]:Image enhancement and image matching are two basic research contents in the field of image processing and computer vision.Most of the traditional image enhancement and image matching methods are based on integral order calculus, which is not ideal for weak edges and weak textures with weak derivative in images.Fractional differential has the advantages of keeping the weak edge and weak texture details in the image while enhancing the image. However, the existing image enhancement methods based on fractional differential need to find the best order manually.In image matching, the combination of SIFT(Scale Invariant Feature transform algorithm and fractional differential algorithm can extract more feature points in fuzzy image and weak texture image, so the accuracy of matching is improved.However, the choice of optimal differential order still needs manual adjustment, which is time-consuming and laborious.Therefore, this paper focuses on these two problems, the specific work is as follows: 1.By analyzing the effect of fractional differential on signal, the differential order adaptive model in fractional differential image enhancement is constructed. The model is based on the inverse tangent function, the gradient information of the image and the local information entropy.Brightness and contrast are independent variables. The relationship between fractional differential order and image local information is established, so that the best order of each pixel in the image can be automatically calculated according to the local characteristic information of the image.The model is applied to image enhancement of fractional differential Tiansi operator.In order to verify the validity of the model in this paper, several texture images in the standard image library are selected for experiments. The experimental results are qualitatively and quantitatively analyzed, and compared with the second-order differential Laplacian operator, Tiansi operator.Both the theoretical analysis and the experimental results show that the model is effective, and the enhancement effect of the gray image can be continuously changed, which is close to the experimental effect of the best fractional differential order, and accords with the human visual perception.An adaptive fractional order SIFT algorithm is proposed for image matching.Based on the Riemann-Liouvilli R-L-based fractional differential, a new fractional differential mask is designed, which is integrated into the SIFT algorithm to extract more precise and effective key points, thus improving the matching accuracy of the SIFT algorithm.Then, according to the local characteristics of the image, a fractional differential order adaptive mathematical model is constructed.In this model, the relationship between fractional differential order and local feature information is established by using the gradient information of image and the entropy of local information as independent variables, using the inverse tangent function as the prototype, and taking the gradient information of image and the entropy of local information as independent variables.In order to verify the validity of the algorithm and model, we can automatically calculate the best order of each pixel in the image according to the local feature information of the image. In order to verify the validity of the algorithm and the model, we choose the image and the real image in the standard library to carry on the experiment.Compared with the original SIFT algorithm and the fractional differential based SIFT algorithm, the efficiency and optimal differential order of the algorithm are analyzed. The theoretical analysis and experimental results show the effectiveness of the proposed algorithm.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】