基于人像检测的实时图像智能裁剪

Real-time automatic image cropping based on portrait detection

针对现有图像智能裁剪算法在处理人像图片时,存在误将主要人物整个人或部分身体部位裁减掉等致使图像关键信息缺失和构图不佳的问题,本文提出一种基于人像检测的实时图像智能裁剪方法。该方法将图像裁剪分为人像检测和智能构图两个阶段,旨在将提取的人像坐标信息作为智能裁剪算法的输入,然后结合基于美学的构图法则对图片进行自动构图,确保裁剪结果中人像信息的完整性并提升图片的构图美感。实验结果表明,本文在Center-Net基础上改造的轻量级检测网络,运算量减少了86%,精度提升了3.34%,便于将该裁剪算法应用于移动端设备。整个裁剪算法的FPS达到了77,并且裁剪后的人像信息完整,图片整体构图得到改善。

In order to solve the problem that the existing automatic image cropping algorithm mistakenly cuts off the whole person or part of the body parts of the main figure,which leads to the missing of the key information of the image and the poor composition of the image,this paper proposes a real-time automatic image cropping method based on portrait detection.This method divides image cropping into two stages:portrait detection and intelligent composition.It aims to take the extracted portrait coordinate information as the input of the intelligent cropping algorithm,and then combines the composition rule based on aesthetics to automatically compose the picture,so as to ensure the integrity of the portrait information in the cropping result and improve the composition beauty of the picture.The experimental results show that the lightweight detection network modified on the basis of Center-Net in this paper reduces the computation amount by 86%and improves the accuracy by 3.34%,which is convenient to apply the cropping algorithm to mobile devices.The FPS of the whole cropping algorithm reaches 77,and the cropped portrait information is complete,and the overall composition of the picture is improved.