The Research Of Digital Image Segmentation And Matting In Image Information Extraction

Digital image segmentation and matting are important technologies in computer vision.They play a major role in image processing and information extraction.There are still some problems worthy for study on image segmentation and matting.How to effectively integrate and utilize the information of different levels and scales in the images in image segmentation to obtain accurate segmentation results and improve detail accuracy.How to better handle highly transparent objects and areas with similar color distribution in foreground and background in image matting.These are very challenging problems in segmentation and matting.For image segmentation,we propose an algorithm named attention based richer convolutional features for image segmentation.This algorithm fully utilizes the rich hierarchical features through the richer convolutional features model.We introduce the spatial grid attention gate to guide the selection of low-level information through highlevel information.Experiments prove that our segmentation algorithm can improve the accuracy of results and the accuracy of details.Our segmentation model makes good use of multi-scale information and selects the most critical information for the segmentation target from the multi-level features.This algorithm has high sensitivity to image features.For image matting,we propose a novel matting algorithm,i.e.adaptive transparency-based propagation matting algorithm.Our matting algorithm assigns the input images into three categories according to the transparencies of the foreground objects in the images.Our matting model can make relevant adjustment in terms of the transparency types of the input images.Our method adds textures as an additional feature to effectively discriminate the foreground and background regions with similar colors.Our matting algorithm gets a good ranking on the benchmark dataset.It provides more accurate results for highly transparent images comparing with the state-of-the-art method.It has a certain degree of discrimination to areas with similar colors.Our method is a new and effective strategy to solve the problem of highly transparent objects.The texture features of the images help to improve the accuracy of the images.