Human image processing method, device and system and readable storage medium

Abstract

The invention relates to a portrait processing method, device and system and a readable storage medium. The method comprises: acquiring a to-be-processed portrait, and generating a plurality of candidate cutting boxes of the to-be-processed portrait; inputting the to-be-processed portrait into the skeleton detection network model for skeleton detection processing to obtain a skeleton node positionof the to-be-processed portrait; calculating a first type of aesthetic quantized value of each candidate cutting frame according to the positions of each candidate cutting frame and the skeleton node; according to each candidate cutting box, cutting the to-be-processed portrait to obtain each candidate cutting image of the to-be-processed portrait; inputting each candidate cutting image into theaesthetic network model to obtain a second-class aesthetic quantized value of each candidate cutting box; and selecting at least one candidate cutting box as a target cutting box of the to-be-processed portrait according to the first type of aesthetic quantized value and / or the second type of aesthetic quantized value of each candidate cutting box. The target cut frame obtained by the method is often accurate.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a portrait processing method, device, system and readable storage medium. Background technique [0002] Based on the pursuit of beauty, people usually crop the image to enhance the composition effect of the image, so that the beauty of the cropped image is increased and the subject is more prominent. [0003] In recent years, many studies have attempted to use deep learning to automatically crop images to be processed. Exemplarily, for the image to be processed, some candidate cropping windows (predefined cropping windows) can be obtained according to different aspect ratios, and each candidate cropping window is obtained by cropping the image to be processed based on each candidate cropping frame; The image features of the candidate cropped images, and according to the image features of each candidate cropped image, calculate the aesthetic score of each candi...

AI Technical Summary

Problems solved by technology

[0004] However, since the above-mentioned deep learning model is generally trained by the loss between the labeled aesthetic score and the predicted aesthetic score of the training sample image, and the labeled aesthetic score has a large subjectivity, the calculation of the above aesthetic score is not reasonable enough. , not stable enough, resulting in the target cropping frame obtained based on the above image cropping method is often not accurate enough

Images