Villa Jesús, Quiroga Juan Antonio, De la Rosa Ismael
Laboratorio de Procesamiento Digital de Señales, Facultad de Ingeniería Eléctrica, Universidad Autónoma de Zacatecas, Avenida Ramón López Velarde 801, 98000 Zacatecas, Mexico.
Opt Lett. 2009 Jun 1;34(11):1741-3. doi: 10.1364/ol.34.001741.
We use the regularization theory in a Bayesian framework to derive a quadratic cost function for denoising fringe patterns. As prior constraints for the regularization problem, we propose a Markov random field model that includes information about the fringe orientation. In our cost function the regularization term imposes constraints to the solution (i.e., the filtered image) to be smooth only along the fringe's tangent direction. In this way as the fringe information and noise are conveniently separated in the frequency space, our technique avoids blurring the fringes. The attractiveness of the proposed filtering method is that the minimization of the cost function can be easily implemented using iterative methods. To show the performance of the proposed technique we present some results obtained by processing simulated and real fringe patterns.
我们在贝叶斯框架中运用正则化理论来推导用于去噪条纹图案的二次代价函数。作为正则化问题的先验约束，我们提出一种马尔可夫随机场模型，该模型包含有关条纹方向的信息。在我们的代价函数中，正则化项对解（即滤波后的图像）施加约束，使其仅沿条纹的切线方向平滑。通过这种方式，由于条纹信息和噪声在频率空间中方便地分离，我们的技术避免了条纹模糊。所提出的滤波方法的吸引人之处在于，使用迭代方法可以轻松实现代价函数的最小化。为了展示所提技术的性能，我们给出了一些通过处理模拟和真实条纹图案获得的结果。
