| With the popularity of digital cameras in the popular, more and more peoplebegin to pay close attention to the merits of the image quality. Because of the variousactual light environment, the no-flash images always appear low brightness, detailsblurring, tone abnormality, etc. and flash images often appear red eyes, flash shadows,strong highlights, poor illumination of distant objects, etc. which damage imagequality. In this thesis, we use the advantages of flash and no-flash images to conductimage details of compensation and image enhancement research.Firstly, we carry on the research of red-eye phenomenon detection and removalalgorithm for flash images. We begin with the classified introduction of existingmethods both at home and abroad. Then we provide our own red-eye detection andremoval method. We detect red-eyes by using skin color detection to narrow the scopeof detection, using red color and saturation characteristics to select the scope, andusing morphological features to exact the detection results. We also correct color andsaturation, remove “dead eye” and smooth image for the red-eye removal and imagequality improvement.Secondly, we commit ourselves to the detection and removal of flashing lightshadow. We compare the edge point information of flash/no-flash image pairs and seta threshold for the difference of gray level image to detect flashing light shadow.After that, we use k-means clustering algorithm to classify the shadow we detect andimport Von Kries chromatic adaptation model to correct umbra region and use linearchange model to correct penumbra region.Finally, we conduct image details of compensation and image enhancementresearch. We choose the color level of flash image as the new image’s color level andcompensate its details by adding the flash image’s high frequency information to theluminance level. We also want to improve the brightness of the whole image. So wefuse flash/no-flash image pairs’ low frequency of luminance level according to certainproportion. |
PDF Full Text Request