Research On Image Denoising Approach Based On High-throughput Genome Sequencing Base Calling

The high-throughput genome sequencing research is a major technology of the second genome sequencing technologies and has attracted tremendous attention from many research institutions. The original image consists of fluorescent spots with base information, and we can get the base of the genome by base calling. The accuracy of the base calling operation highly relies on the quality of the original fluorescent image, but a fluorescent image has a low signal-to-noise ratio(SNR). Hence preconditioning on the raw fluorescent image is crucial to the high-throughput genome sequencing technology. The research on the sequencing image denoising algorithm has made a great process in the last few decades, and a category of wavelet thresholding denoising shrinkage algorithms(WTDSA) has a great performance on solving sequencing images noising problems, due to the sequencing images are distinguished for their diverse texture density and noise. This dissertation proposed two improve robust wavelet thresholding denoising algorithms for sequencing image denoising.There are two categories in WTDSA traditionally:The wavelet denoising shrinkage algorithm based on global thresholding and the wavelet denoising shrinkage algorithm based on local thresholding. This dissertation first proposes a wavelet denoising improvement algorithm based on à trous wavelet. The noised image is decomposed as several high-frequency components and a low-frequency component. Using the 1 norm to construct the global thresholding expression at each level, we can get the estimator wavelet coefficients by the novel expressions which are associated with the wavelet coefficients and thresholding. In the end, we can get the denoising image by the inverse wavelet transform. The second method in the dissertation is the improved algorithm of the Bi Shrink thresholding based on the discrete wavelet transform。The traditional Bishrink algorithm calculates the estimator coefficients which take account of the correlation between the current coefficients and parent-child coefficients. The new algorithm selects some coefficients in the neighborhood to calculate the local threshold at each level of wavelet coefficients, and improves the estimator expression at each level by Laplace model,to perform robust sequencing image denoising on the datasets with Gaussian noise.Experiments on two high-throughput sequencing databases, including the Swift and synthetic database, have been conducted to test the frameworks proposed in this dissertation. Meanwhile, some recently proposed powerful WTDSA are utilized for comparisons. We use the SNR and mean square error(MSE) as metrics of evaluation. The experimental results demonstrated that the proposed frameworks have high denoising performance and also can dramatically improve the robustness.