<![CDATA[Image completion, inpainting or hole-filling, is the process to fill or replace an area of image with another image so the replacement can not be detected. The replacement area is the area from the image that is similar than image that has the removing area. The feature of color image that can be used is statistic feature, such as mean, median, and standard deviation. Template matching is used to find the area from the similar image. Then, image blending, like Poisson Blending, is used to equalize gradient of the area from the similar image. The image dataset is obtained from LabelMe and the resolution of images is 256x256. The number of training image is 10 images and 3 images as testing images or query images. The accuracy to produce the scene completion is 66.67%.]]>
Copyrights © 2015
