【正文】 的灰度平均值代替原來點(diǎn)的灰度值，此算法經(jīng)多次疊代可增強(qiáng)平滑效果，消除噪聲的同時，又可很好地保持邊沿的銳度。在每次實(shí)驗(yàn)中 ,把不同密度的椒鹽噪聲隨機(jī)地加到原始圖像上。 根據(jù)圖像統(tǒng)計特性 ,按某一函數(shù)關(guān)系加權(quán)到各個元素， 且兩邊元素的權(quán)重正比于方差的大小 。 進(jìn)行加權(quán)時 ,可取排序后窗口中的幾個元素(如 33 窗口 ,取其中 3個元素 )進(jìn)行加權(quán)。 figure,imshow(K3) （ a） （ b） （ c） （ d） 圖 21 調(diào)入函數(shù) medfilt2 實(shí)現(xiàn)中值濾波 ， （ a）加入噪聲圖像 , （ b） 3*3 的中值濾波后結(jié)果 （ c） 5*5的中值濾波后結(jié)果 , （ d） 7*7 的中值濾波后結(jié)果 從上圖中可以看出，中值濾波器不像均值濾波器那樣，它在衰減噪聲的同時不會使邊界模糊。% 進(jìn)行 3*3中值濾波 K2=medfilt2(I,[5,5])。 pepper39。39。例如 ： Med(1,3,4,0,6)=3。這里的鄰域稱為窗口 ,當(dāng)窗口在圖像中上下左右進(jìn)行移動后 ,利用中值濾波算法就可以很好地對圖像進(jìn)行平滑處理。中值濾波器在濾除噪聲的同時能很好地保護(hù)圖像邊緣 ,使圖像較好地復(fù)原。子波收縮 (硬閉值或軟閉值 )其實(shí)就是一個二元判決，當(dāng)子波系數(shù)大于一定閉值時，保留該系數(shù)，反之予以置零，然后用新得到的系數(shù)重構(gòu)信號。 小波降噪問題，前人已經(jīng)做了不少工作 : 和 Hwang 通過計算 Lipschitz 正則來刻劃信號的奇異性，將小Lipschitz 指數(shù)的子波系數(shù)去掉，用剩余系數(shù) 重構(gòu)信號達(dá)到去噪的目的。 關(guān)鍵詞： 圖像去噪 ， 維納濾波，中值濾波，小波變換，閾值 Abstract In its formation, transmission and recording of the process of digital images, because imaging system , transmission media and recording equipment are often imperfect, the obtained images are polluted by a variety of noises. In pattern recognition, puter vision, image analysis and video coding and other fields,noise image preprocessing is extremely important and whether its effect is good or bad will have a direct impact on the following quality and results. This paper introduces the basic principle, the typical method and the latest methods of image the rapid development of technology of image denoising into account, the paper discusses the basic theory and at the same time also the latest research results and the latest methods in recent years. This paper is divided into four The first part is the introduction and discusses development trend of image denoising and the reasons and significance of studying image denoising. The second part, deals with the basic principles of median filter and adaptive smoothing filter, achieves the pletion of median filtering code based on Matlab, and analyzes the results. This paper presents two new algorithm, which is the improved algorithms of the filtering called adaptive weighted algorithm, and the improved algorithm of adaptive smoothing. And the paper has reached this algorithm simulation results, and analyzed the results. The third part firstly discusses the basic principles of image denoising based on frequency domain . Then this paper discusses the basic principles of Butterworth lowpass filter and Butterworth highpass filtering, and pletes the code achieved based on Matlab Butterworth lowpass filter and highpass filtering and analyzes the results. Meanwhile important statements of the procedures are explained. The fourth part of this article is the most important chapter and focuses on the two methods and algorithms of image denoising based on wavelet domain, which are the wavelet domain thresholding method and wavelet wiener filter method. In wavelet thresholding method, the paper focuses on the three steps of wavelet thresholding and discusses the traditional classical threshold methods,which are soft, and the threshold hard threshold law, and introduces four ways of determining the four ways include a single threshold value, interval threshold based on the zero mean normal confidence, the largest minimum threshold value and ideal threshold paper pletes achieving code of wavelet thresholding method and paratively analyzes the results of wavelet thresholding method and the results of denoising filter method. In wavelet wiener filter ,the paper method focuses on the basic principle of wavelet wiener filter, achieves simulation results of wavelet wiener filter method, and pares the results of wavelet wiener filter method with the results of the wiener filter method. Keywords : image denoising, Wiener filter, filtering, wavelet transform, threshold 第 1 章 緒論 圖像去噪的發(fā)展趨勢 圖像信號處理中最困難的問題之一是 :怎樣濾出圖像中的噪聲而又不模糊圖像的特征及邊緣。第四部分是本文最重要的一章， 重點(diǎn)闡述 基 于小波域的兩種圖像去噪方法 和算法，即小波閾值去噪法與小波維納濾波去噪法。本文提出 兩種新 的 算法，即中值濾波的改進(jìn)算法 即 自適應(yīng)加權(quán)算法 ， 和 自適應(yīng)平滑濾波的改進(jìn)算法。考慮到圖像去噪技術(shù)的飛速發(fā)展，本文在論述其基本理論的同時還著重介紹近年來國內(nèi)有關(guān)的最新研究成果和最新方法。摘 要 數(shù)字圖像在其形成、傳輸和記錄的過程中，由于成像系統(tǒng)、傳輸介質(zhì)和記錄設(shè)備的不完善往往使得獲取的圖像受到多種噪 聲的污染。 本文 被分成 四個部分。并且也得出 這兩種算法 的仿真結(jié)果，并且對 結(jié)果 進(jìn)行分析。在小波閾值去噪法中，本文重點(diǎn)論述 小波閾值去噪的三個步驟， 并介紹 傳統(tǒng)經(jīng)典的閾值化方法即軟閾值法 、硬閾值法以及四種確定閾值的方法。傳統(tǒng)的圖像去噪方法有兩類 :一類是頻率域方法，另一類是空間域方法。Mallat 的方法實(shí)現(xiàn)復(fù)雜，對于低信噪比信號去噪效果并不好。 [1] 研究圖像去噪的理由與意義