Micro-focus X-ray Image Denoising Based On Sparse Representaion

With the development of the integrated circuit(IC) manufacturing technology, the chips are becoming more integrated, the electronic component types are more and more rich, which proposes higher requirements for detecting internal defects of IC packaging. X-ray inspection as a non-destructive testing technology, plays an important role in the IC packaging detection. Due to X-ray focal spot size is inversely proportional with the imaging resolution, it is necessary to use micro-focus X-ray tube in order to meet the requirements for IC packaging precision detection.The focal spot size of micro-focus X-ray tube is only microns, so X-ray flux is little per unit time which results in micro-focus X-ray images with low signal-to-noise ratio and serious mixed noise. Meanwhile, IC usually consists of copper, silicon and metal sheet, the difference of thickness is small, the material types are limited. In the process of X-ray imaging, X-ray energy attenuation in different areas differs little, leading to low-contrast micro-focus X-ray images. Traditional image denoising methods can’t obtain good results in micro-focus X-ray images. Image denoising methods based on sparse representation aim at restoring images’ sparsity. They have good image processing results and they are very suitable for micro-focus X-ray images that are sparse naturally. Therefore, the study of this paper concentrates on micro-focus X-ray image denoising method based on sparse representation. The main content and research results in this paper are as follows:1. From the relationship between noise and image, the complex noise existing in micro-focus X-ray image is modeled as mixed additive and multiplicative noise. Firstly the model is decomposed appropriately and the additive and multiplicative noise is removed step by step. Details are as follows: the objective function removing additive noise is constructed based on the principle of total variation and the objective function removing multiplicative noise is proposed based on sparse representation of wavelet analytic dictionary. Then the additive noise and the multiplicative noise are filtered by explicit difference method and gradient projection in steps. Finally, from subjective visual assessment and evaluation of objective indicators, the mixed additive and multiplicative noise model is verified the effectiveness and the experiments demonstrate that the proposed method can not only denoise the micro-focus X-ray image effectively but also preserve the image details well.2. The image model is established to represent the micro-focus X-ray images with mixed poisson and gaussian noise from the point of different noise models. According to this model, the denoising method based on sparse representation and dictionary learning is proposed, which adopts the OMP algorithm for sparse coding and K-SVD method for adaptively dictionary updating. And the poisson data fidelity term is approximated through the second-order Taylor approximation. Finally, the experiments on micro-focus X-ray images show that the poisson and gaussian mixture model is correct and the proposed method is effective. Furthermore, the experiments on standard images demonstrate that the proposed method has a certain range of applicability and can effectively remove the mixed poisson and gaussian noise.