Underwater Image Restoration by Red-Dark Channel Prior and Point Spread Function Deconvolution

在水下研究領域裡，無人載具通常都會安裝影像監控系統來記錄海底環境資訊。其拍攝的水下影像擁有些共同特徵：色偏和低能見度。這些因素是由於光進入海水之中，其能量會以指數性衰減；此外，根據波長的性質，衰減的幅度也會有所不同。此篇論文主要以簡化的傑夫-葛萊蒙利(Jaffe-McGlamery)水下光學模型來有效率地修復水下影像。在我們的方法裡，首先利用改良過的暗黑頻道來估算紅光色系的背景光和穿透率。如此一來，其能見度將會以物體與相機之間的相對距離獲得適當的補償。接者，藉由分析點擴散方程式與前向散射的關係，我們推導出一個簡單但實用的低通濾波來對水下影像去捲積。最後，利用整張影像的灰階值與波長吸收係數之間的關係，估測出場景距海平面的平均深度及對各色系做合理地補償色差。實驗的數據指出，在各種水下場景，提出的演算法會比許多現存的方法擁有更高程度的影像品質。因此，我們深信此修復方法能夠延伸出不同的水下影像之應用層面。In the field of undersea research, underwater vehicles usually carry camera systems for recording. The captured images and videos often have two undesired characteristics: color distortion and low visibility. This is because that the light is exponentially attenuated while penetrating through water. Furthermore, the quantity of attenuation is associated with the wavelength of light spectrum. This thesis simplifies the Jaffe-McGlamery optical model and proposes an effective algorithm to recover underwater images. In our approach, a red-dark channel prior was defined and derived to estimate the background light and the transmission. The visibility of scene was compensated by the object-camera distance to recover the color of the background and objects. Subsequently, by analyzing the physical property of the point spread function, we developed a simple but efficient low-pass filter to deblur the image by deconvolution. Finally, we used the relationship between the radiance above the sea surface and the absorption coefficient to correct the color distortion. A wide variety of underwater images with different scenarios were used for the experiments. The experimental results show that the proposed algorithm effectively recovered underwater images while eliminating the influence of absorption and scattering. Comparing with many existing methods, the proposed framework provided more natural restoration results. We believe that this new restoration algorithm is promising in many underwater image processing applications