一种新的基于自适应分数阶的活动轮廓模型

发布时间：2018-05-05 18:46

本文选题：Grünwald-Letnikov分数阶 + 自适应分数阶　；参考：《计算机研究与发展》2017年05期

【摘要】：区域可调拟合(region scalable fitting,RSF)活动轮廓模型在分割弱纹理、弱边缘图像时,优化易陷入局部极小导致曲线演化速度缓慢;同时该模型中的局部拟合项为高斯核函数,导致目标的边界模糊,影响分割精度.针对该问题,提出了一种基于自适应分数阶的活动轮廓模型,用于图像的分割.首先将全局G-L(Grünwald-Letnikov)分数阶梯度融合到RSF模型中,以增强灰度不均匀和弱纹理区域的梯度信息,从而提高对曲线初始位置选择的鲁棒性,并提高了图像分割的精度和速度;然后用双边滤波函数替换局部拟合项中的高斯核函数,解决了高斯核函数在演化过程中造成的边界模糊问题;最后根据图像的梯度模值和信息熵构建自适应分数阶阶次的数学模型,并计算出最佳分数阶阶次.理论分析和实验结果均表明:提出的算法可以用于灰度不均匀和弱纹理、弱边缘区域的图像分割,并能根据图像的特征自适应计算最佳分数阶阶次,避免曲线演化陷入局部最优.用多幅图像进行实验,得出该方法的分割精度和分割效率都有较大提高.
[Abstract]:The active contour model of region scalable fitting (RSF) can segment weak texture and weak edge image, and the local fitting term in the model is Gao Si kernel function, and the curve evolvement speed is slow due to its easy to fall into local minima, and the local fitting term in the model is Gao Si kernel function, and the local fitting term of the model is Gao Si kernel function. The boundary of the target is blurred and the segmentation accuracy is affected. To solve this problem, an adaptive fractional-order active contour model is proposed for image segmentation. Firstly, the global G-L(Gr 眉 nwald-Letnikov-fractional-step degree is fused into the RSF model to enhance the gradient information of the inhomogeneous and weakly textured regions, so as to improve the robustness of the initial position selection of the curve and the accuracy and speed of image segmentation. Then, the Gao Si kernel function in the local fitting term is replaced by the two-sided filter function, which solves the boundary ambiguity problem caused by Gao Si kernel function in the evolution process. Finally, a mathematical model of adaptive fractional order is constructed according to the gradient modulus and information entropy of the image, and the optimal fractional order is calculated. The theoretical analysis and experimental results show that the proposed algorithm can be used for image segmentation in the regions of uneven grayscale, weak texture and weak edge, and can adaptively calculate the best fractional order according to the characteristics of the image. Avoid curve evolution falling into local optimum. Experiments with multiple images show that the segmentation accuracy and efficiency of the method are greatly improved.
【作者单位】：江西省图像处理与模式识别重点实验室(南昌航空大学);西华大学机械工程学院;
【基金】：国家自然科学基金项目(61462065) 江西省自然科学基金项目(20151BAB207036) 江西省科技支撑计划基金项目(20161BBF60091)~~
【分类号】：TP391.41

【相似文献】