Adaptive Regularization Denoising Methods For Micro-focus X-ray Images

X-ray imaging is an effective means to detect the internal defects in integrated circuit(IC) packaging components. The detective objects in IC like thin wires, tiny solder joints are much smaller than those in industry or medical diagnosis. To meet the image resolution, it’s essential to use the X-ray tubes with focal spot less than 6μm.However, the smaller X-ray tube’s focal spot, the less X-ray flux emitting per unit time, resulting in a low level photon count which will cause a noisy image. Meanwhile, packaging components consist of silicon wafers, foils, connecting wires and so on. During imaging, X-ray penetrates them and attenuates rarely, leading to a low-contrast X-ray image. Because of X-ray image’s characteristic, i.e. small features, serious noisy and low contrast, traditional restoration algorithms for a particular noise model can’t achieve good smoothing results for micro-focus X-ray images. On the other hand, regularization methods are flexibility to any noise models and adaptive regularization methods can preserve edges simultaneously smooth noise. So this thesis concentrates on adaptive regularization restoration methods for micro-focus X-ray images. Following are the main works and achievements of this dissertation.(1) The complex noise existing in micro-focus X-ray images is modeled as mixed Poisson-Gaussian noise based on X-ray imaging mechanism. Then the total variation denoising model and adaptive regularization denoising model for mixed noise are attained under the maximum a posteriori estimator. Experimental results show the mixed noise model is more reasonable than single noise model, such as Gaussian noise or Poisson noise.(2) Due to X-ray images’ characteristics and traditional filtering methods’ shortcomings, an adaptive regularization denoising method based on phase congruency is proposed, which applies different regularization operators on different regions to remove noise and retain edges.(3) Given the disadvantage of phase congruency when it extracts the contours near image boundaries, a dual adaptive regularization method based on local variance is put forward. It utilizes local variance to divide the image into two parts, namely homogeneous regions and detail regions. Then an isotropic regularization operator and a regularization parameter with a big value are applied on homogeneous regions, while an anisotropic operator and a regularization parameter which is inversely proportional to the local variance adopted on detail regions. Numerical simulations demonstrate the proposed method is not only able to smooth the mixed noise, but also capable to preserve edges and details, which greatly improves the peak signal-to-noise ratio and image quality.