實現在手機平台之無鬼影高動態範圍影像合成技術

Ghost-free High Dynamic Range Imaging on Mobile Phone

李侑倫

指導教授 ： 江瑞秋

國立中正大學/工學院/電機工程學系/碩士(2016年)

摘要

相信大家都有為了保存美好回憶而拿起手邊的相機拍照的經驗，一般人拍照不會像專業攝影師隨身攜帶腳架，避免不了因為拍照時，拍攝者手持相機造成晃動或拍攝過程中場景內有物體移動，導致照片模糊。這些移動都會毀掉一張相片，對於喜愛拿著相機到處拍照的人而言，這將是個非常困擾的事情。
高動態範圍 (High Dynamic Range, HDR) 影像，通常是用多張不同曝光照片，透過軟體合成，呈現亮暗部曝光細節皆清楚的影像，較能貼切描述真實世界中，從太陽光直射到最暗陰影的亮度範圍。我們提出的無鬼影高動態範圍影像合成演算法主要針對合成時出現的模糊區域進行探討，解決鬼影問題，適用於手持式裝置晃動與物體移動產生的模糊情境。我們拍攝兩張不同曝光影像進行HDR合成，短曝光影像稱為I_s(E_0,view0)，長曝光影像稱為I_l(E_1,view1)，由於拍攝時可能手持相機造成晃動或拍攝過程中場景內有物體移動，因此，I_s與I_l我們設定為不同視角(view0,view1)，若直接進行HDR合成，會出現一張有模糊區域的影像，嚴重影響影像品質。故我們提出創立一張I_t^'(E_1,view0)擁有I_s視角資訊(view0)和I_l曝光亮度(E_1)，進行無鬼影HDR合成，即可得到一張高品質無鬼影的HDR影像。另外，傳統HDR採用pixel-based計算權重，我們提出使用patch-based的權重計算方式，加快HDR合成速度。
我們將提出的無鬼影patch-based HDR演算法實現在Android平台上，搭載在ASUS ZE551ML硬體裝置上。常用HDR合成在手機上運行要花費4秒時間，但存在鬼影問題，而我們的無鬼影HDR合成花費約6秒合成時間，解決鬼影問題，讓喜愛使用手機拍照的民眾也可以享受無鬼影HDR帶來的高品質照片。

關鍵字

高動態範圍 ； 手持式裝置 ； 無鬼影HDR合成 ； Android平台 ； patch-based HDR

並列摘要

Cameras have a limited dynamic range, much smaller than that of most scenes, or even the human eye. This means low-dynamic-range (LDR) images captured by cameras may lose detail in the very dark or bright areas of the scene. Several hardware and software techniques have been developed in order to capture the entire visual information present in a high-dynamic-range (HDR) scene. Ghosting artifacts are usually caused by moving object when composing a high dynamic range image from multiple differently exposed conventional images. We present an effective fusion technique for an exposure bracketed images, which may contain motion blur from moving objects or hand trembling. To have a high quality result without resorting to image alignment, we propose to fuse a histogram-transformed image I_t^'(E_1,view0) with the short-exposed image to preserve desired properties. Our approach eliminates the blur and ghosting artifacts by fusing the input images. We create two versions of the final image: one noisy but unaffected by motion blur I_s(E_0,view0), and one potentially blurry but colorful and less affected by noise I_l(E_1,view1). We fused the noisy and blurred images to preserve their desired qualities in the final picture. Traditional HDR is used pixel-based to compute weights, we proposed to use the right patch-based weight calculation to accelerate HDR synthesis rate. We will propose a ghost-free patch-based HDR algorithms implemented on the Android platform, so that people like to use a mobile phone camera can also enjoy no ghosting HDR to bring high-quality photos.

並列關鍵字

LDR ； HDR ； patch-based HDR ； Histogram-transformed ； Android

參考文獻

[2] Greg. Ward, “The Log Luv Encoding for Full Gamut High Dynamic Range Images”, Journal of Graphics Tools, Vol. 3, No. 1, pp.15-31, 1998.

[4] P. E. Debevec and J. Malik, “Recovering High Dynamic Range Radiance Maps from Photographs,” Proc. SIGGRAPH 97 conference Proceedings, Annual Conference Series, pp.369–378, August 1997.

[5] T. Mertens, J. Kautz, and F. Van Reeth, “Exposure Fusion,” Proc. Pacific Conf. on Computer Graphics and Applications, pp. 382-390, Jan 2007.

[6] S. Raman and S. Chaudhuri, "Bilateral filter based compositing for variable exposure photography", Proc. Eurographics, pp. 1-4, 2009.

[7] M. Song, D. Tao, C. Chen , J. Bu and C. Zhang, "Probabilistic Exposure Fusion", IEEE Transactions on Image Processing., Vol. 21, No. 1, pp. 341 -357, 2012.