In this paper, we proposed a deep learning based automated colorization method using stepwise classification to improving the coloring application rate of the colorization method. The image problem of the stained color of the result of VGG-16 coloring is applying to the convolution layers 5 and 6 by applying the Inception V2 classifier in two stages. To solve stained problem of contour area, we proposed an Automated Colorization method using a stepwise. The existed colorization method is a method of coloring through the auto encoder, caused a lot of feature loss. This is an unnatural result because the coloring of the outline of the object is processed in black or white during the coloring process, and the coloring of the image shows the color of the stained form as a whole is difficult to color. By using the proposed method, VGG-16 feature extraction is relatively clear compared to auto encoder method, which is a compression method, and low level feature and high level feature are computed by applied a stepwise for consistent coloring. Experiments with nature, structures, and objects show that the proposed method has a 9.4% improvement in coloring application rate.
Volume 11 | 06-Special Issue
Pages: 1974-1980