“Automatic Deblurring for Facial Image Based on Patch Synthesis” by Kawai, Iwao, Maejima and Morishima

  • ©Masahide Kawai, Tomoyori Iwao, Akinobu Maejima, and Shigeo Morishima



Entry Number: 16


    Automatic Deblurring for Facial Image Based on Patch Synthesis



    Deblurring of a facial image is one of the most important topics in computer vision. Blurriness can occasionally appear on a facial image in different parts, such as a eyes, a nose, a cheek and a mouth because of alignment errors. For example, the morphable model and active appearance model are statistical models that can represent a variety of facial images by changing parameters; however, synthesized facial images can include the same type of blurs as mentioned previously.
    To solve these problems, Li et al. [2010] proposed a robust motion deblurring system. Their method enables deblurring of the flat cheek region of facial images, but it cannot fix inconsistent images, which are defined as images that include considerably stronger blurs, such as the eyes and the nose. For example, there are strange blurs, such as the white of an eyeball tinged with black color and the nasal cavity tinged with skin color because facial images are created by morphing each pixel. Therefore, we propose a novel method to automatically deblur a variety of facial images that include inconsistent images. Our method is considerably more effective in strong blurs because it can create a novel image by replacing a blurred image with small square images called “patches,” without additional databases.


    1. Li, X., and Jiaya, J., 2010. Two-Phase Kernel Estimation for Robust Motion Deblurring, 11th ECCV, pp. 157–170.
    2. Mohammed, U., Prince, J. D. -S., and Kautz, J., 2009. Visio-lization: Generating Novel Facial Images, SIGGRAPH’09, pp. 57(1)–57(8).


ACM Digital Library Publication:

Overview Page: