Research On Fine-Grained Semantic Segmentation And Colorization For National Costume Grayscale Images

Grayscale image colorization has always been a popular research topic in the field of computer vision,in many aspects such as digital image restoration has a great demand for application scenarios.As an important technique for the digital protection and inheritance of ethnic culture,the colorization of grayscale image of national costumes can effectively play the application of color elements of national costumes in industrial design education and other fields,which has certain research significance and value.However,due to the rich and complex color characteristics and diverse color distribution of national costumes,automatic colorization of national costume grayscale images have become a challenging task.In the traditional mainstream shading task,the image is usually treated as a whole object and the semantic information difference of different regions is ignored.In order to solve the limitation of traditional mainstream colorization methods in the application of grayscale image coloring of national costumes,consider each part of the national costume color effect of semantic information,in this paper,This paper introduces fine-grained semantic information,semantic for national costumes of grayscale images segmentation and automatic colorization method was studied,the main research contents and results are as follows:(1)The fine-grained semantic high-definition image dataset of national costumes is constructed.In order to carry out the research on semantic segmentation and colorization of grayscale images of national costume,this paper constructs a fine-grained semantic dataset of national costumes containing more than 1200high-definition clothing images in 8 semantic categories of Dai,Hani,Wa and Yi.(2)The semantic segmentation model of national costume grayscale images were designed.The model and loss function were further optimized by embedding the semantic segmentation model based on the context perception of edge by the way of human body parsing model.The experimental results show that the model designed in this paper can obtain more accurate fine-grained semantic segmentation results for the gray images of national costumes.(3)A fine-grained semantic automatic colorization model for national costume grayscale images were proposed.The model was improved on the Pix2 Pix HD network model,which uses fine-grained semantic segmentation results and original grayscale images for feature connection,and as the input condition of conditional generative adversarial networks,uses costume semantic information for auxiliary coloring.The experimental comparison shows that this model can solve the problems of color distribution correspondence and spatial consistency of costume image well,and has a good performance in the task of national costume grayscale image colorization.(4)The fine-grained semantic colorization system of national costume is designed and implemented.Based on Py Torch deep learning frame,the semantic automatic segmentation,artificial semantic correction,grayscale image colorization and other functions are integrated,and a fine-grained semantic segmentation and colorization system for gray image is designed and implemented.