The Research And Design Of Image Coloring Based On Deep Learning

Grayscale image coloring is an important research content of image processing system.Image coloring is widely used in the restoration of old photos,color restoration of cultural relics and film and television processing.The original coloring technology has two methods: local color extension based on artificial coloring and color transfer based on reference image.However,due to the large proportion of human influence of these two methods,the coloring effect is not very good and cannot be used for a large number of images.With more and more developed science and technology,due to the popularization and rapid development of artificial intelligence technology and intelligent hardware and software equipment,people are constantly considering to intelligentize all kinds of mechanized and programmed equipment and automatically solve some subjective,repetitive and complicated tasks by using relevant artificial intelligence technology.In the environment of artificial intelligence,the rapid development of computer vision also follows.It is also imperative to combine deep learning with image processing.This paper analyzes and experiments several methods of image coloring comparison based on deep learning:(1)According to the original image color need to think to participate in,and the effect difference obvious drawbacks,the original generated against the defect of network,the condition is adopted to generate the image color against network and after it was found that,although conditions generated against network can generate color images,but not controllable,lead to the result is not the best;(2)An image translation algorithm is proposed,which is improved on the basis of generating the antagonistic network and solves the uncontrollability of generating the antagonistic network.(3)Image translation is suitable for coloring in the case of large data sets,so a transfer learning algorithm is proposed for the case of small data sets.(4)Therefore,this paper proves the effectiveness and significance of deep learning algorithm through theory and practice.