Research And Implementation On Image Content Changes Based On Deep Feature Interpolation

High-confidence semantic editing of images is an important research direction in computer graphics,and it is also a hot research field at present.With the rapid development of information technology,image data is increasing rapidly,and has become a very important information medium.Image semantic editing technology plays a decisive role in many fields such as scientific research,advertising design,film and television production,animation materials,etc.Even in the case of public security criminal investigation,it also puts forward an urgent need for the technology,so the study of image semantic editing technology has very important meaning.Deep learning,especially convolutional neural networks,has made great achievements in many image fields with its excellent design ideas and excellent performance.Based on the advantages of deep learning theory and practice,this paper designs and implements an image transformation algorithm model based on depth feature interpolation to perform high-confidence semantic editing for multiple attributes of face images.The main work of this article is as follows:1.The idea of depth feature interpolation is that the steps of image editing are done in the depth feature space,not in pixel space.This paper designs and trains a convolutional neural network,which is used for image feature extraction as a function of image depth feature mapping.2.Considering that in the deep representation space of the neural network,the fusion effect between different models is better,and more detailed image information is captured in the shallow feature space of the neural network,which is conducive to the reconstruction of the image in the pixel space.In the training network,a dense connection module was introduced to extract image feature information,and the shallow and deep information flow of the neural network was comprehensively considered.3.Depth features need to map to pixel space to generate digital images.For this task we designed a new objective function,which introduced the depth image priori as a priori term in order to reconstruct a pixel image.4.Designed and implemented the removal of color noise and modified traces in face images,improving the quality of the edited renderings.In the experiment,we tested the face attribute editing algorithm on the Labeled Faces in the Wild database,and compared the quality of the editing effects with the other two algorithm models.Experimental results show that the image transformation method used in this paper has a good performance in the credibility and quality of editing.