Research And Application Of Video Colorization Method Based On Deep Learnin

Computer vision has become an important area of artificial intelligence with the development of the times.Its purpose is to use cameras and computers to simulate human vision to perceive and understand things in the outside world by a specific method,to achieve a set purpose.Colorization,as a significant branch of computer vision,aims to restore images and videos with missing or corrupted colors,thus satisfying people’s need for vision,and color information also plays a significant role in other vision fields,therefore,the research of colorization methods is crucial.The problems faced in the study of video colorization methods are mainly colorization quality and temporal consistency.For the colorization quality problem,the aim is to take a given input grayscale frame and,through a series of methods and operations,recover its color information so as to output color frames.Temporal consistency,on the other hand,effectively ensures that the video color frames can be color-smooth between them without flickering and local color mutations.In this regard,these main issues have been thoroughly analyzed and discussed,and the following research work has been carried out.1.Combining Cross-scale Fusion and Recursion for Video Colorization Adversarial Generative Networks is proposedFor the temporal consistency problem,we utilize the network structure of RNN,as the model shape of RNN does not change with the input length,and the calculation takes into account the history information to ensure the degree of correlation between successive frames,thus better ensuring the continuity between frames so that the consecutive frames get good temporal consistency in colorization;for the colorization quality problem,we utilize the UNet++ coding method to decode the video frames,integrating different levels of features to improve accuracy,and then enhance the object boundary information in the video frames,thus improving the colorization quality of grayscale video frames.In addition,we innovatively utilize the style transformation loss function in terms of images for model training to ensure temporal consistency and colorization quality by exploiting the slight difference in style variation from frame to frame.2.Proposed Transformer-based Video Colorization MethodWe use the existing image colorization method as our colorization module for pre-colorization the video once to guarantee the colorization quality of the video grayscale frames,and we propose to apply Transformer to achieve temporal consistency in video colorization,making full use of local information through the self-attentive layer and using the motion information of the bidirectional optical flow to discover the correlation between different video frames and perform feature fusion to ensure better results in terms of timing.3.Design and implementation of a video colorization system based on deep neural network algorithmThe system mainly designs and implements a video colorization system based on the deep neural network algorithm based on the above algorithm.And the system mainly consists of three parts,which are login/registration,video coloring processing and coloring effect display,among which video coloring processing provides the coloring method selection.After users register or log in to the system through the Web terminal,they select the coloring method they need,then color the video,and finally display the effect of the colored video.