Algorithm Of Fundus Images' Denoising And Enhancing Based On NSCT

Retinal images are important for diagnosing diseases related to the eye and of great values to the modern medical. However, because of the limitations of the eye's structure, retinal images are always with various edge details but have low contrast and nonuniform illumination. It brings lots of inconvenience for doctors'diagnosis and for the fundus images'further processing. The traditional denoising and enhancing algorithms often improve the contrast and simultaneously blur image detail characteristics, which is unacceptable for medical image processing that requires images of higher quality. Multiscal geometric analysis, as a breakthrough of the multiscale analysis, which not only inherits the multiresolution characteristics and the time-frequency localization properties, but also is anisotropic and multidirectional, can provide excellent tools for medical image processing.After studying the principles and structures of Contourlet transform and Non-Subsampled Contourlet Transform (NSCT), and analyzing the characteristics of retinal images and various algrithms of denoising and enhancing, the paper proposes a methord combined denoising and enhancing for improving fundus images based on NSCT, which uses the Principle Component Analysis (PCA) to select the threshold for denoising. Firstly, the image is decomposed into coefficient arrays of different scales and different directions using NSCT, then noise energy is estimated directly by PCA. After the selection of the threshold and the adjusting parameter according to the noise energy estimated, coefficient arrays are modified in accordance with the threshold function, and the final enhancing result is acquired by the inverse transform.The paper mainly adopts the objective evaluation to evaluate the algorithm, together with the subjective evaluation. The Peak Signal to Noise Ratio (PSNR) is used to evaluate the effect of denoising and the method of Background Variance and Detail Variance (BV-DV) is for the enhancement. The results of experiments show that compared with other methods, the proposed method can increase PSNR greatly, and also improve the image contrast and the visual effect.