用於相機的數位影像處理

Abstract

近年來，影像與視訊相關技術及商品蓬勃發展，衍生了許多與生活息息相關的應用，除了應用於醫學、天文、交通監控、安全監控、工業檢測等應用外，消費性應用更是眾所矚目的焦點，如電影工業、(數位)相機、(數位)攝影機、具備相機模組的行動通訊設備、玩具等。 數位影像處理技術在上述應用中，扮演了舉足輕重的角色。 本研究提供了下述五個問題的解決方案，包含從光學鏡頭生產線上就會遇到的瑕疵檢測問題 (Lens Shading Correction for Dirt Detection)，用於立體相機的一致性前處理 (Conformity Preprocessing for 3D Stereo Camera)，以及提供了三個影像處理管線 (Image Processing Pipeline) 之後的數位影像處理技術，包含可用於提高人物拍攝品質的自動膚色美化技術 (Automatic Skin Color Beautification)、可用於提高影像擷取設備景深涵蓋範圍的影像合成技術 (Clear Focused Image from Macro and Infinite Images)、可用於突顯拍攝主題的離焦放大數位影像後處理技術 (Defocus Magnification with CUDA)。 由本研究所提供的實驗結果可看出，應用於光學鏡頭髒污、瑕疵檢測的技術能有良好的檢出能力。在確認無光學鏡頭髒污影響之後，應用我們提出的前處理技術於相機影像處理管線中，能夠達到提高相機色彩一致性的特性。至此，相機硬體的問題將有效獲得改善，同時，本研究也針對近年熱門的影像處理應用提出建議與演算法。自動膚色美化技術能應用於不同膚色人種、不同環境色溫下拍攝的膚質美化。影像合成技術能在影像對位完美且無相機放大效應的假設前提下，透過近景對焦與無限景深對焦的兩張圖片融合出景深涵蓋較大的影像。由離焦放大之影像後處理技術的實驗結果能夠看出，我們的處理效果相較於實驗對照組來說，在細節處理上表現的更優異。

Recently, the application of image and video grows massively in multiple fields, including medical science, astronomy, traffic application, surveillance system, industrial inspection, and so on. In addition, the applications on consumer electronics are closely related to our daily life. For example: film industry, digital still camera, digital video camcorder, mobile device or toy with camera module, and so on. Digital image processing technology plays an important role in abovementioned applications. In our research, we propose a solution to help automatically detect defect optical lens from production lines: Lens Shading Correction for Dirt Detection. Besides, we propose Conformity Preprocessing for 3D Stereo Camera to improve the color consistency between different devices. Next, we provide an Automatic Skin Color Beautification technology to embellish the portrait image automatically. Moreover, we propose an image processing technology to fuse macro and infinite images to extend the depth-of-field: Clear Focused Image from Macro and Infinite Images. Finally, we provide an improved post-processing technology (Defocus Magnification with CUDA) to make the subject in photographs more prominent. The Experimental result of Lens Shading Correction for Dirt Detection shows that our proposed method performs outstandingly well. By obviating the effects of imperfect optical designs, the proposed Conformity Preprocessing for 3D Stereo Camera shows that the color consistency between different devices can be improved significantly. Moreover, we propose three digital image processing approaches to practical camera applications. The results of the proposed Automatic Skin Color Beautification technology shows that it can be applied in various light environments, and different kinds of skin colors. Moreover, the results of our Clear Focused Image from Macro and Infinite Images technology can extend the depth-of-field in the condition which has good image registration and no camera magnification effect. Lastly, the results of Defocus Magnification technology show that our method can generate more continuous defocus map, and that the processed results are more uniform than Bae’s results.