本文選題：隨機(jī)脈沖噪聲 + 閾值敏感性　； 參考：《中國(guó)圖象圖形學(xué)報(bào)》2017年06期

【摘要】：目的隨機(jī)噪聲的噪聲閾值具有不確定性和敏感性,尋找一個(gè)魯棒的閥值是非常困難的,這嚴(yán)重影響了噪聲的提取效率。為提高噪聲判斷的準(zhǔn)確性,提出一種基于方向特性與中智不確定性融合的雙端脈沖檢測(cè)算法;另外,為加強(qiáng)優(yōu)良像素在濾波過(guò)程中的權(quán)重,構(gòu)建了一種基于像素中智不確定性和ROAD(rank-ordered absolute differences)統(tǒng)計(jì)量的新型雙邊濾波函數(shù)。方法在噪聲檢測(cè)階段,首先根據(jù)ROLD(rank-ordered logarithmic difference)與噪聲閾值T的關(guān)系,將污染圖像的像素分為超限域像素(ROLD≥T)、鄰限域像素(0.8T≤ROLDT)和安全域像素(ROLD0.8T),并利用開關(guān)機(jī)制完成一次噪聲檢測(cè)。在此基礎(chǔ)上,為提高超限域和鄰限域像素噪聲檢測(cè)的準(zhǔn)確性,采用不同策略對(duì)其進(jìn)行二次噪聲排查:對(duì)超限域像素,利用新型25像素和9像素4方向模板計(jì)算像素基于排序的方向?qū)?shù)差統(tǒng)計(jì)量,由該統(tǒng)計(jì)量與T的大小關(guān)系決定當(dāng)前像素的噪聲真?zhèn)?對(duì)鄰限域像素,則結(jié)合當(dāng)前像素中智不確定性在濾波窗內(nèi)的排序信息來(lái)進(jìn)一步確定其噪聲特性。在濾波階段,利用像素中智不確定性和ROAD統(tǒng)計(jì)量構(gòu)建新型雙邊濾波函數(shù),以加強(qiáng)低不確定性和高相似性像素在圖像恢復(fù)中的權(quán)重。結(jié)果針對(duì)實(shí)驗(yàn)圖像,雙端脈沖檢測(cè)算法的邊緣像素提取率最高可達(dá)67%、鄰限域像素的噪聲剔除率最高可達(dá)91%,大大降低了閾值對(duì)噪聲提取的敏感性,從而提高了噪聲判斷的正確率。在10%~80%噪聲范圍內(nèi),本文算法的主觀性能和峰值信噪比都優(yōu)于其他7種算法。結(jié)論本文基于雙端檢測(cè)和新型雙邊濾波函數(shù)的新算法,在噪聲檢測(cè)和去噪過(guò)程中均充分考慮了圖像本身的方向性和噪聲的不確定性,因此提高了噪聲提取及像素濾波權(quán)重的準(zhǔn)確性,從而有效地保護(hù)了圖像的邊緣和細(xì)節(jié)信息。
[Abstract]:Objective the noise threshold of random noise is uncertain and sensitive, so it is very difficult to find a robust threshold, which seriously affects the efficiency of noise extraction. In order to improve the accuracy of noise judgment, a two-terminal pulse detection algorithm based on the fusion of directionality and uncertainty is proposed, and the weight of fine pixels in the filtering process is enhanced. A new two-sided filter function based on pixel intelligence uncertainty and ROAD(rank-ordered absolute differences statistics is constructed. Methods in the phase of noise detection, according to the relationship between ROLD(rank-ordered logarithmic difference) and noise threshold T, the pixels of contaminated image were divided into two categories: the pixel in the over-limited domain (r), the pixel in the adjacent domain (0.8T 鈮，

本文編號(hào)：1945590