强激光与粒子束, 2013, 25(3): 751, 网络出版: 2013-03-01
Inpainting method for flash radiographic anti-scatter grid image based on neural networks
中国工程物理研究院 流体物理研究所, 四川 绵阳 621900
提出了一种基于径向基函数(RBF)神经网络的闪光照相网栅图像修补算法, 该方法采用滑动窗口方法将待修补的网栅图像分为若干子块, 然后在每个子图像内分别引入RBF神经网络, 将栅孔内图像作为已知数据计算RBF网络参数, 并以此对每个子图像进行修补, 数值试验表明, 该算法能较好地再现图像边缘信息, 修复的图像在信噪比和视觉方面都优于线性插值和样条插值的结果。
To solve the problem of flash radiographic anti-scatter grid image inpainting, a radial basis function (RBF) neural network based image inpainting algorithm is proposed. First the anti-scatter grid image is divided into a series of blocked images. Then the weights of the RBF network are estimated and a continuous function is constructed in each blocked image, and with them the pixels of missing information can be filled in. The experimental results show that the new algorithm has better general performance in inpainting quality and boundary maintenance compared with the linear interpolation and spline interpolation method.