Research And Implementation On Image Colorization Algorithm Based On Deep Learning

It is an important part of the image processing system to colorize the grey-scale map.As a kind of computer-aided technology,image coloring has a wide range of applications in the field of video processing and old photo restoration.The current image colorization techniques mainly based on color transfer and color-based expansion of two methods.First of all,it requires human interaction and professional image processing that is difficult to process.Moreover,the coloring effects is poor.Besides,it can't be extended to a large number of different types of images according to a coloring algorithm,which makes it difficult to be applied to industrial production in large quantities.With the rapid development of technologies such as digital image processing,computer vision,machine learning and deep learning,deep learning and digital image processing are becoming more and more integrated.The image processing methods based on deep learning also become diversified.Therefore,the image coloring algorithm based on deep learning has a very important meaning.Aiming at many influencing factors of image coloring and deep learning,this paper has done the following research:Adopting the public deep learning image dataset,including animals and plants,mountains,forests and other complex and varied,multi-angle images.In order to overcome the drawbacks of the traditional coloring algorithms such as human intervention and poor coloring effect,this paper proposed an image coloring algorithm based on residual neural network that is compared with the coloring algorithms in recent years.The test proves that the proposed image coloring algorithm based on residual neural network is better than other coloring algorithms.Proposed an image coloring algorithm based on generative adversarial networks(GANs).Completed the image coloring system based on deep learning that consists of image upload module,image preprocessing module and image coloring module.The image coloring module uses the two image coloring algorithms proposed by this paper,which can colorize a large quantity images at once.Through the combination of theory and practice,this paper proves the validity and significance of the proposed algorithm.