The technique of laminography is used to image different planes of interest in an object. A laminograph is obtained by shifting and aligning several radiographs. Boundaries of features at different focal planes are used for alignment. Hence the degree of accuracy of alignment depends on better edge or boundary detection of features. Since real radiographs always contain noise, therefore some kind of noise removal technique also has to be employed. The most common noise removal procedure of low-pass filtering results in the loss of important edge information. Hence to form a precise laminograph a noise reduction technique is required that reduces noise as well as preserves, if not enhances, edges. Nonlinear diffusion filtering provides such a technique. Nonlinear diffusion filtering is the solution of the nonlinear diffusion equation in which the diffusion coefficient is chosen such that it minimizes diffusion across the edges hence preserving them while the diffusion process in the interior of regions reduces noise. Usually, the numerical implementation of the nonlinear diffusion equation is done using explicit schemes. Such schemes impose strict restrictions on the time step-size for stability, and hence require numerous iterations, which leads to poor efficiency. If the semi-implicit scheme is used for the numerical solution of the nonlinear diffusion equation, the results are good and stable for all time step-sizes. The application of the nonlinear diffusion equation using the semi-implicit scheme on a radiograph results in noise reduction and edge enhancement, which in turn means a more precise laminograph. The effect of this technique on real images is shown in comparison to the conventional methods of noise reduction and edge detection. © 2000 American Institute of Physics. [ABSTRACT FROM AUTHOR]