對於汙染文字影像分割之基於U-Net的空洞卷積神經網路架構

U-Net based Atrous Convolutional Neural Network Architecture for Polluted Text Image Segmentation

李吉恩(JI-EN LI)

指導教授 ： 饒建奇

淡江大學/工學院/電機工程學系碩士班/碩士(2022年)

https://doi.org/10.6846/TKU.2022.00386

摘要

由於紙本保存文獻資料的方式在現今環境氣候變化越來越激烈的情況下，
紙本保存文獻資料的方式越來越不容易，紙張受到潮濕得影響容易泛黃使閱讀困難，文字染料的部分也容易因此產生污漬，使得重要資訊無法被辨識，且紙張也容易受外力保存不當而毀損。
本篇碩士論文展示了如何應用神經網路架構的方式，在已經遭受汙染的文字影像中，如何把重要的文字資訊切割、保存起來，如此一來重要的文字資訊便可以轉換成電子檔保存起來，使得重要的文字資訊可以不再受氣候環境的影響，不再使得文字閱讀困難，本篇碩士論文也運用了一個跟傳統卷積不一樣的做法，使得文字影像分割出來之後，對於人眼辨識的效果可以有所提升，降低了周圍雜訊對於人類眼睛辨識力的干擾。

關鍵字

神經網路架構 ； 空洞卷積 ； U-Net

並列摘要

Due to the way in which documents are preserved on paper, in the context of today's increasingly drastic changes in the environment and climate, It is becoming more and more difficult to preserve documents on paper. The paper is affected by moisture and is prone to yellowing, making it difficult to read. The dyed part of the text is also prone to stains, making important information unrecognizable, and paper is also prone to improper preservation by external forces and damaged.
This paper shows how to use the neural network architecture to cut and save important text information, so that important text information can not be affected by the climate and environment, and it will no longer make the text difficult to read. This paper also uses a different from the traditional convolution method, after the text image is segmented, the effect of human eye recognition can be improved, and the interference of surrounding noise on human eye recognition can be reduced.

並列關鍵字

Neural Network Architecture ； Atrous convolution ； U-Net

參考文獻

[1] Liang-Chieh Chen, George Papandreou, Florian Schroff, Hartwig Adam, “Rethinking Atrous Convolution for Semantic Image Segmentation” the International Conference on Medical Image Computer Assisted Intervention(),

[2] Olaf Ronneberger, Philipp Fischer, and Thomas Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation” the International Conference on Medical Image Computer Assisted

Intervention(MICCAI),2015

[3] Evan Shelhamer, Jonathan Long, Trevor Darrell, “ Fully Convolutional Networks for Semantic Segmentation” the International Conference on Medical Image Computer Assisted Intervention(CVPR),2015

[4] Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, Hartwig Adam, “ Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation” the International Conference on Medical Image Computer Assisted Intervention(ECCV),2018