An Improved BM3D Method for eDNA Mieroarray Image Denoising

Abstract

Microarray arouse tremendous attentions owing to its outstanding advantages on dealing with tens of thousands genes simultaneously, and image processing is a crucial step in microarray analysis. However, real images are usually obtained according to a series of procedures, which will bring noises or artifacts that result in poor image quality. To improve the image processing accuracy, noise reduction is crucial. Therefore, in this paper, we introduce the state-of-art method, block-matching and 3D filtering algorithm, into microarray image denoising. First, median filter and contrast enhancement method are added into image preprocessing to improve the image quality. Next, the threshold and Wiener contraction coefficient involved in initial and final estimation are improved according to image variance. Experiments on real images drawn from the SMD, GEO, BCM, DeRisi and SIB datasets indicating that the proposed method perform better compared to the Donoho threshold, generalized wavelet, adaptive thresholding, compressed sensing, non-local means methods. Quantitative analysis on 127 sub-grids also verifies the efficiency of our proposed method.