Color Image Sketch Stylization with Extended Difference of Gaussian (XDoG) and Deep Learning
Journal
Journal of Information Science and Engineering
Journal Volume
40
Journal Issue
3
Start Page
437
End Page
453
Date Issued
2024-05-01
Author(s)
Chen, Chia-Yi
DOI
10.6688/JISE.202405_40(3).0001
Abstract
Artists paint portraits using sparse lines and shaded areas to capture the unique look of a face. The face photo sketches generated by deep learning techniques and eXtended Difference of Gaussian (XDoG) edge detection operator are monotonous since they only have two-tone color results, black and white. Therefore, we dedicated to developing an algorithm of color face photo sketch based on existing technology. In this paper, the XDoG operator and its extensions are explained in detail, and its parameters are arranged. Furthermore, we demonstrate the application of the XDoG operator, and render the proposed artistic effect of the pencil sketch style and display the technical details and stylistic potential of this operator. Moreover, we apply our proposed color face sketch method not only to the retinex theory, but also to the distinct color spaces (HSV, Lab, and YCbCr), and then analyze and discuss these differences. Through the experimental results, the differences between the color face photo sketch based on the XDoG operator and the deep learning are compared and the visual quality obtained by the method based on the XDoG is better than the system based on the deep learning. Besides, our proposed color face photo sketch can effectively generate high-quality, expressive, and color artistic face drawings from face photos whether for animals or human. Finally, the organized and expanded XDoG will have more practicality and commercial value.
Type
journal article