臺灣大學: 資訊網路與多媒體研究所李明穗蔡佳娜Tsai, Chia-NaChia-NaTsai2013-03-222018-07-052013-03-222018-07-052011http://ntur.lib.ntu.edu.tw//handle/246246/251200許多消費性電子產品或是電腦視覺的應用上，例如：自動化監控系統、倒車導航等，單張影像除霧被賦予極高的關注。當行經多霧路段時，對所擷取的影像移除所覆蓋的霧氣，不僅提升視覺能見度，也讓行車過程中更加安全。 從物體成像的角度而言，進入人眼的是一道無色且複合的能量。人眼所感受到的色彩，是由於環境中的物體吸收部分入射光能後，反射剩餘能量，傳遞至人眼所致。在光線的傳遞過程中，能量會因為環境中的懸浮粒子而散射流失，形成能量的直接衰減(direct attenuation)。然而，當入射光直接撞擊於這些細微的懸浮粒子時，產生能量完全散射。這些被空氣中的懸浮微粒所散射的懸浮散射光(airlight)覆蓋於物體所反射的能量之上，導致進入人眼的能量，經過視網膜上的錐狀細胞以及大腦神經的交互作用下產生霧濛濛的影像。為了除去不必要的懸浮散射光，還原真實的物體反射能量，本篇論文提出了根據不同色光(color component)的穿透能力(transmittance)來實現單張影像除霧。在提出的方法中，使用了以物理為基礎的圖片降階模型，搭配光學理論來幫助提升影像品質。此圖片降階模型為反射能量的直接衰減以及懸浮散射光的線性組合。 根據光學理論，能量散射程度與色光波長成反比。換言之，所得到的不同色光能量與其波長成正比，即波長愈長，所保留的能量愈多。因此，分割已知能量(energy separation)來模擬不同色光經傳遞後所得的光能；以及基於被霧所覆蓋之影像特性，利用對比最大化(contrast maximization)估計出環境中的懸浮散射光來推導不同色光的穿透能力，進而達成霧化影像的復原。 實驗的結果顯示，提出的方法能有效地除去圖片中因霧化所造成的影響且改善圖片細節。此外，透過與其他目前先進的單張影像除霧的方法比較，來探討所提出之方法的執行成果。Haze removal is highly desired in both computer vision applications and consumer photography. The presence of suspended particles in the atmosphere disturbs the transference of light and leads direct attenuation to a reflectance of scene radiance. When the incident light hit directly to these tiny particles, the scattered light called airlight which is covering over the reflectance of scene radiance and reaching human’s eyes or cameras forming a hazy scene or a foggy image. The haze-covered scene or images might lose details and have the color gloomy. To address this problem, an effective single image dehazing method according to the transmittance of different light components is proposed in this novel. A physics-based image degradation model of a linear combination of direct attenuation and airlight adding with optics theory are utilized in the proposed method. In this thesis, we assume that the arrived light is a type of energy which is compounded and achromatic. The color that we felt is the product of an interaction between cones in the retina and nerves of the brain. Estimating the airlight by computing a suitable value which can produce a maximum contrast. After the derivation of airlight, transmission maps are straightforward. Recovery of scene radiance is easily achieved by above steps. Experimental results of proposed method demonstrate its ability to recover the scene radiance and repair image details. The performance of proposed method is further explored via comparative studies with several state-of–the-art single image dehazing methods.3312106 bytesapplication/pdfen-US單張影像除霧能量分割對比最大化懸浮散射光估計色光穿透能力反射能量回復Single Image DehazingEnergy SeparationContrast MaximizationAirlight EstimationTransmittance of Light ComponentsRadiance Recoverythesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/251200/1/ntu-100-R98944023-1.pdf