Unsupervised Deep Image Stitching: Reconstructing Stitched Features to Images

图像拼接 人工智能 计算机科学 计算机视觉 像素 无监督学习 特征提取 深度学习 特征(语言学) 模式识别(心理学) 图像分辨率 迭代重建 语言学 哲学
作者
Lang Nie,Chunyu Lin,Kang Liao,Shuaicheng Liu,Yao Zhao
出处
期刊:IEEE transactions on image processing [Institute of Electrical and Electronics Engineers]
卷期号:30: 6184-6197 被引量:188
标识
DOI:10.1109/tip.2021.3092828
摘要

Traditional feature-based image stitching technologies rely heavily on feature detection quality, often failing to stitch images with few features or low resolution. The learning-based image stitching solutions are rarely studied due to the lack of labeled data, making the supervised methods unreliable. To address the above limitations, we propose an unsupervised deep image stitching framework consisting of two stages: unsupervised coarse image alignment and unsupervised image reconstruction. In the first stage, we design an ablation-based loss to constrain an unsupervised homography network, which is more suitable for large-baseline scenes. Moreover, a transformer layer is introduced to warp the input images in the stitching-domain space. In the second stage, motivated by the insight that the misalignments in pixel-level can be eliminated to a certain extent in feature-level, we design an unsupervised image reconstruction network to eliminate the artifacts from features to pixels. Specifically, the reconstruction network can be implemented by a low-resolution deformation branch and a high-resolution refined branch, learning the deformation rules of image stitching and enhancing the resolution simultaneously. To establish an evaluation benchmark and train the learning framework, a comprehensive real-world image dataset for unsupervised deep image stitching is presented and released. Extensive experiments well demonstrate the superiority of our method over other state-of-the-art solutions. Even compared with the supervised solutions, our image stitching quality is still preferred by users.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
李爱国应助姜有明采纳,获得10
刚刚
sheng完成签到 ,获得积分10
1秒前
梓渝完成签到 ,获得积分10
1秒前
Ruby于完成签到 ,获得积分10
2秒前
2秒前
2秒前
淡定的幼晴完成签到,获得积分10
3秒前
Ca0cus完成签到,获得积分10
3秒前
4秒前
王顺发完成签到 ,获得积分10
4秒前
崔康佳完成签到,获得积分10
7秒前
cao发布了新的文献求助10
7秒前
猪猪hero发布了新的文献求助10
7秒前
文静的翠彤完成签到 ,获得积分10
8秒前
俞孤风完成签到,获得积分10
9秒前
jun完成签到,获得积分10
10秒前
10秒前
Sampson完成签到,获得积分10
10秒前
禹宛白发布了新的文献求助10
11秒前
13秒前
在水一方应助猪猪hero采纳,获得10
13秒前
yyy完成签到 ,获得积分10
13秒前
JamesPei应助猪猪hero采纳,获得10
13秒前
wanci应助默默毛豆采纳,获得10
16秒前
小蘑菇应助zy0411采纳,获得10
16秒前
姜有明发布了新的文献求助10
16秒前
xumq发布了新的文献求助10
17秒前
闪闪小蜜蜂完成签到,获得积分10
17秒前
贰鸟发布了新的文献求助200
19秒前
薄荷心完成签到 ,获得积分10
22秒前
bkagyin应助Menghang采纳,获得10
23秒前
23秒前
李响完成签到,获得积分10
25秒前
大雪完成签到 ,获得积分10
25秒前
YongLiu完成签到,获得积分10
26秒前
小雨唱片完成签到,获得积分10
26秒前
li应助ii采纳,获得20
27秒前
猪猪hero发布了新的文献求助10
27秒前
蓓蓓完成签到 ,获得积分10
27秒前
28秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
Organic Reactions, Volume 116 1500
VALIDATION OF THE TAYLOR, ALAMEL AND VPSC MODELS FOR PLASTIC ANISOTROPY MODELING OF SHEET METALS 1000
Geist der Kunst und Kultur 1000
Middleton's Allergy Principles and Practice 10th Edition(Middleton's Allergy 2-Volume Set, 10th Edition) 1000
Resistance Spot Welding Dataset for Automobile Body-in-White Quality Analysis 748
日本現代怪異事典 副読本 700
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 工程类 有机化学 化学工程 生物化学 计算机科学 内科学 物理 复合材料 催化作用 细胞生物学 无机化学 光电子学 物理化学 电极 基因
热门帖子
关注 科研通微信公众号,转发送积分 7401535
求助须知:如何正确求助?哪些是违规求助? 9006310
关于积分的说明 19172390
捐赠科研通 7035362
什么是DOI,文献DOI怎么找? 3231104
关于科研通互助平台的介绍 2393416
邀请新用户注册赠送积分活动 2212836