人工智能培训

搜索

人工智能论文:一种用于场景文本识别的多目标纠正注意网络(A Multi-Object Rectified Attention Network for Scene

[复制链接]
rrrrrrr 发表于 2019-1-11 10:59:50 | 显示全部楼层 |阅读模式
rrrrrrr 2019-1-11 10:59:50 280 0 显示全部楼层
人工智能论文:一种用于场景文本识别的多目标纠正注意网络(A Multi-Object Rectified Attention Network for Scene Text Recognition)不规则文本被广泛使用。然而,由于其各种形状和扭曲的图案,因此很难识别。因此,在本文中,我们提出了一种用于一般文本识别的多目标整流关注网络(MORAN)。 MORAN由多目标整流网络和基于注意力的序列识别网络组成。多目标整定网络用于纠正包含不规则文本的图像。它降低了识别的难度,并使基于注意的序列识别网络更容易读取不规则文本。它以弱监督方式进行训练,因此只需要图像和相应的文本标签。基于注意力的序列识别网络主要关注目标特征,并依次输出预测结果。此外,为了提高基于注意力的序列识别网络的灵敏度,提出了一种基于训练阶段的基于训练的解码器的分数拾取方法。通过整改机制,MORAN可以读取常规和不规则的场景文本。在各种基准测试中进行了广泛的实验,这表明MORANachieves具有最先进的性能。源代码可用。
Irregular text is widely used.However, it is considerably difficult torecognize because of its various shapes and distorted patterns.In this paper,we thus propose a multi-object rectified attention network (MORAN) for generalscene text recognition.The MORAN consists of a multi-object rectificationnetwork and an attention-based sequence recognition network.The multi-objectrectification network is designed for rectifying images that contain irregulartext.It decreases the difficulty of recognition and enables theattention-based sequence recognition network to more easily read irregulartext.It is trained in a weak supervision way, thus requiring only images andcorresponding text labels.The attention-based sequence recognition networkfocuses on target characters and sequentially outputs the predictions.Moreover, to improve the sensitivity of the attention-based sequencerecognition network, a fractional pickup method is proposed for anattention-based decoder in the training phase.With the rectificationmechanism, the MORAN can read both regular and irregular scene text.Extensiveexperiments on various benchmarks are conducted, which show that the MORANachieves state-of-the-art performance.The source code is available.人工智能论文:一种用于场景文本识别的多目标纠正注意网络(A Multi-Object Rectified Attention Network for Scene Text Recognition) NBn6wb5EeNgPAFNF.jpg
URL地址:https://arxiv.org/abs/1901.03003     ----pdf下载地址:https://arxiv.org/pdf/1901.03003    ----人工智能论文:一种用于场景文本识别的多目标纠正注意网络(A Multi-Object Rectified Attention Network for Scene Text Recognition)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

rrrrrrr当前离线
新手上路

查看:280 | 回复:0

快速回复 返回顶部 返回列表