Automatic image editing method

Abstract

The invention discloses an automatic image editing method. The method comprises the following steps: constructing a first full convolutional neural network model for labeling a tripartite graph of a foreground, a background and an uncertain region corresponding to the photo based on an encoder-decoder; obtaining a first original picture to perform image processing transformation, and generating acorresponding template picture as first training data; enabling the first full convolutional neural network model to firstly perform random rotation, scaling and gamma transformation processing on theoriginal image according to the first training data, combining the original image and the alignment template image into an image in a channel direction, and outputting a real tripartite graph; constructing a second full convolutional neural network model for realizing image matting based on an encoder-decoder; obtaining a second original picture for image processing and synthesis, and generatingan RGB picture as second training data; and enabling the second full convolutional neural network model to perform random rotation, scaling and gamma transformation processing on the RGB image, combining the original image and the real tripartite graph in the channel direction to form an image, and outputting a matted image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an automatic image editing method. Background technique [0002] With the wide coverage of technological applications in modern life, various wearable devices and smart identification device applications can be seen everywhere in our lives. Image information such as photos and videos is playing an increasingly important role in people's lives. People are beginning to get used to recording their lives with cameras, and Weibo and Moments are full of photos taken by everyone. The emergence of massive image information also means that there is a higher demand for image information processing. Thousands of videos and images are transmitted on the Internet every day. Technical terms such as video tracking, image detection, and image analysis are no longer limited to advanced scientific fields, but have also penetrated into people's daily life with the maturity and promotion o...

AI Technical Summary

Problems solved by technology

[0004] The present invention provides an automatic image editing method to solve the technical problem of poor effect of the existing matting processing technology, thereby completing precise matting by constructing a fully convolutional neural network model with two encoder-decoder structures, Then realize the optimization of the matting technology and improve the processing effect

Images