【摘要】：室外圖像往往會(huì)因?yàn)榭諝庵袘腋☆w粒(霧霾顆粒等)的存在而產(chǎn)生圖像降質(zhì)的問(wèn)題。而降質(zhì)問(wèn)題會(huì)嚴(yán)重影響室外視頻監(jiān)控及圖像采集系統(tǒng)的工作效率,因此一直以來(lái),研究者們都致力于除霧算法的研究來(lái)提升這些系統(tǒng)的魯棒性及可靠性。另一方面,霧的存在是圖像保持空間透視距離感的關(guān)鍵組成,研究者們對(duì)如何在CG場(chǎng)景中產(chǎn)生逼真的煙霧環(huán)境產(chǎn)生了濃厚的興趣,煙霧等環(huán)境同時(shí)也能產(chǎn)生水墨畫等的藝術(shù)效果。然而遺憾的是,目前卻沒(méi)有工作將除霧以及霧效模擬結(jié)合,而二者本質(zhì)核心都是對(duì)透射率進(jìn)行計(jì)算。因此,本文提出了面向有霧圖像的透射率模型,同時(shí)可進(jìn)行除霧及霧效模擬。算法首先對(duì)基本的大氣散射模型進(jìn)行轉(zhuǎn)化,引入了最大能見(jiàn)度,將有霧圖像之間聯(lián)系起來(lái),進(jìn)而提出了新的自然透射率模型(霧效濾波模型),并輔以顏色校正和天空補(bǔ)償來(lái)提升算法效果。對(duì)于算法核心的圖像透射率估計(jì),本文提出了兩種不同的技術(shù)路線：一是根據(jù)單一暗原色先驗(yàn)知識(shí)進(jìn)行估計(jì),并使用暗原色快速算法,引導(dǎo)濾波進(jìn)行加速,這種方法操作簡(jiǎn)單,計(jì)算迅速,保證了算法運(yùn)算的效率,運(yùn)算速度可達(dá)到10-1s數(shù)量級(jí),基本達(dá)到實(shí)時(shí)性要求；二是引入了高斯過(guò)程回歸算法,利用有霧圖像及對(duì)應(yīng)透射率的訓(xùn)練集,以透射率作為輸出向量,利用多尺度多維度的特征向量學(xué)習(xí)獲得有霧圖像透射率,這種方法具有更高的通用性與適用性。實(shí)驗(yàn)結(jié)果表明,我的方法所得到的除霧圖像具有良好的清晰度,真實(shí)度,對(duì)比度,與此同時(shí)霧效模擬的結(jié)果真實(shí)而自然。
[Abstract]:Outdoor images tend to degrade due to the presence of suspended particles (haze particles) in the air. However, the degradation problem will seriously affect the efficiency of outdoor video surveillance and image acquisition systems, so researchers have been working on de-fogging algorithm to improve the robustness and reliability of these systems. On the other hand, the presence of fog is a key component of the image's sense of distance from perspective. Researchers are interested in how to create realistic smog environments in CG scenes. The environment such as smoke can also produce the artistic effect of ink painting and so on. However, unfortunately, there is no work to combine fog removal and fog effect simulation, and the core of both is to calculate the transmittance. Therefore, a transmittance model for foggy images is proposed, which can be used to simulate fog removal and fog effect. The algorithm firstly transforms the basic atmospheric scattering model, and introduces the maximum visibility to link the fog images. Furthermore, a new natural transmittance model (fog effect filter model) is proposed, which is supplemented by color correction and sky compensation to improve the effectiveness of the algorithm. For the image transmittance estimation of the core of the algorithm, this paper proposes two different technical routes: one is to estimate the image transmittance according to the prior knowledge of single dark primary color, and to use the fast dark primary color algorithm to speed up the image transmission estimation by guiding filter. This method is simple to operate. The calculation is rapid, which ensures the efficiency of the algorithm, and the operation speed can reach the order of 10 ~ (-1) s, which basically meets the real-time requirements. Secondly, Gao Si's process regression algorithm is introduced, which makes use of the fog image and the corresponding transmittance training set. The transmittance is used as the output vector and the multi-scale and multi-dimensional eigenvector is used to obtain the transmittance of foggy image. This method is more general and applicable. The experimental results show that the defogging images obtained by my method have good sharpness, fidelity and contrast, while the simulation results of fog effect are real and natural.
【學(xué)位授予單位】：大連理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類號(hào)】：TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 郭t，

本文編號(hào)：2282689