Comparative Study of Tongue Image Denoising Methods

Abstract

Tongue diagnosis is one of the most important examinations in traditional Chinese medicine. Tongue images are often corrupted by various noises, but the subsequent diagnosis requires that the tongue images are clean, clear and noise-free. Thus, tongue image denoising is the vital preprocessing step in tongue diagnosis. A comparative study of tongue image denoising methods is given in this work, and four different methods, i.e. wavelet transform, wavelet packet transform, adaptive median filter and wiener filter are utilized to denoise the tongue images, and then the performance of these four methods is evaluated and compared. The experimental results show that wavelet transform-based method can effectively reduce Gaussian noise and speckle noise mixing in the tongue images and yields better results than the other three methods, and the adaptive median filter method gets the best result in removing salt & pepper noise. Moreover, the results also indicate that wavelet transform-based method outperforms wavelet packet transform-based method for the noise reduction of tongue images.