影像處理中，影像去雜訊是個非常重要的研究課題，近年來，許多專家和學者已經提出非常多效果不錯的去噪方法。在實際應用上，可以利用在古畫去污、耳機的即時降噪、生物醫學影像的去噪等。然而，大部分的方法只適用於處理影像中只有一種雜訊的情形。事實上，在一些真實圖像中經常含有多個雜訊組成的混合噪聲。例如螢光顯微鏡的成像中就含有泊松和高斯的混合雜訊。此外，混合雜訊在去噪過程上也較困難。在這篇文章中，我們首先介紹 P-M 模型和分數階層模型是如何去除椒鹽雜訊。同時我們應用小波閥值、非局部平均模型和三維塊匹配算法去除高斯雜訊。接著提出我們的演算法去除混合雜訊和加入總變差去噪的加速法來縮短模型的運算時間，此外也介紹 Noise2Noise 模型的即時去噪法去除混合雜訊。最後我們展示每種方法在對各種混合雜訊進行去噪時的效果和對這些實作結果做總體比較分析。;Image denoising is one of important research topics in image processing. In recent years, many scholars and experts have proposed many effective denoising methods. In practical applications, it can be used for removing dirt from ancient paintings, instant noise reduction of headphones, denoising of biomedical images, etc. However, most methods are suitable while the image contains only one type of noises. In fact, some real images often contain mixed noise composed of multiple noises. For example, the images of fluorescence microscopy contain mixed noises of Poisson and Gaussian noise. In addition, mixed noise is also more difficult to denoise. This paper first introduces how the respective P-M model and fractional order model remove salt and pepper noise. We simultaneously apply wavelet thresholding, NL-Means model and BM3D algorithm to remove Gaussian noise. Secondly, we propose our algorithm for removing mixed noise and add the accelerated method of total variation denoising to shorten the running time of the model. Furthermore, we introduce the real-time denoising method of the Noise2Noise model for removing various mixed noises. Finally, we show the effect of each method on denoising various mixed noises and make an overall comparative analysis of these implementation results.
