Lightweight image automatic clipping system and method based on deep reinforcement learning

Abstract

The invention discloses a lightweight image automatic clipping system and method based on deep reinforcement learning. The system comprises an environment, an action space and an agent embedded in theenvironment, wherein the environment provides current observation for the intelligent agent, calculates an actual reward value of a cutting action, executes the cutting action on a current observation image, and updates the current observation; the intelligent agent comprises a pre-trained convolutional neural network MobileNetV2 model and two full connection layers and is used for extracting image features and outputting a clipping action value and an estimated state value. The action space provides an actual cutting action for the environment according to the cutting action value output bythe intelligent body; the intelligent agent can automatically learn how to make a sequence clipping action, an IOU value calculated through the environment is used as a reward function, and SOTA performance can be achieved with fewer clipping steps and shorter clipping time.

Description

technical field [0001] The invention belongs to the technical field of image processing, and more specifically relates to a lightweight image automatic cropping system and method based on deep reinforcement learning. Background technique [0002] With the continuous increase of the current image data volume, the demand for automatic image processing is gradually increasing, and image cropping is a very important step in image processing. Image automatic cropping technology can not only quickly complete the processing of most pictures, but also assist professional photographers to find a better perspective to improve the quality of image composition, which has great application value. [0003] In the past, most image cropping tools need to generate a large number of candidate cropping windows from the input image, and then select the most beautiful image from the large number of candidate cropping windows as the final cropped image. This process not only consumes calculation ...

AI Technical Summary

Problems solved by technology

[0006] Aiming at at least one defect or improvement requirement of the prior art, the present invention provides a lightweight image automatic cropping system and method based on deep reinforcement learning, which regards the automatic image cropping process as a sequence decision process and an agent-environment interaction problem , its agent automatically learns how to make sequence cropping actions during the training process, and uses the average IOU value calculated by the environment as part of the reward function. The purpose is to solve the existing image cropping methods that consume large computing resources and Time, the problem of not being able to accurately obtain high-quality cropped images

Images