首页--工业技术论文--自动化技术、计算机技术论文--计算技术、计算机技术论文--计算机的应用论文--信息处理(信息加工)论文--模式识别与装置论文

Deep Learning Based Urdu Optical Character Recognition

摘要第5-7页
ABSTRACT第7-8页
Biographical Sketch第9-21页
CHAPTER 1 Optical Character Recognition第21-44页
    1.1 Background of Optical Character Recognition (OCR)第21-25页
    1.2 Urdu Language第25-27页
        1.2.1 Urdu Character Set第25页
        1.2.2 Diacritics第25-26页
        1.2.3 Writing Styles第26-27页
    1.3 Complexities of Nastaleeq Urdu Text第27-33页
    1.4 Segmentation of Urdu Nastaleeq text第33-34页
    1.5 OCRs for Urdu Language第34-40页
        1.5.1 Segmentation based Approach第35-39页
        1.5.2 Segmentation Free OCR第39-40页
    1.6 Motivation第40-41页
    1.7 Contribution of Thesis第41-42页
    1.8 Thesis Structure第42-44页
CHAPTER 2 Preliminaries第44-64页
    2.1 Projection Profile Method第44-45页
    2.2 Salt-and-Pepper Noise (SP)第45页
    2.3 X-Y Cut第45-47页
    2.4 Smearing Algorithm第47页
    2.5 Rectified Linear Unit (ReLU)第47-48页
    2.6 Softmax Function第48页
    2.7 Artificial Neural Network (ANN)第48-54页
        2.7.1 Common Activation Functions第49-50页
        2.7.2 Network Topologies第50-51页
        2.7.3 Learning Paradigms第51-52页
        2.7.4 Perceptron第52-53页
        2.7.5 Multilayer Perceptron第53-54页
    2.8 Autoencoder第54-56页
        2.8.1 Denoising Autoencoder第55页
        2.8.2 Stacked Denoising Autoencoder第55-56页
    2.9 Hidden Morkov Model第56页
    2.10 Support Vector Machines第56-57页
    2.11 Recurrent Neural Network第57-58页
    2.12 Long Short Term Memory第58-60页
        2.12.1 Bidirectional Long Short Term Memory第59-60页
    2.13 Connectionist Temporal Classification第60页
    2.14 Datasets第60-61页
    2.15 Project第61-62页
    2.16 Accuracy Measurement第62页
    2.17 Summary第62-64页
CHAPTER 3 Segmentation for Printed Urdu Text第64-96页
    3.1 Introduction to Segmentation and Challenges for Robust Urdu OCRs第64-66页
    3.2 Related work第66-70页
        3.2.1 Line segmentation第66-68页
        3.2.2 Ligature Segmentation第68-69页
        3.2.3 Motivation第69-70页
        3.2.4 Contribution第70页
    3.3 Line Segmentation第70-81页
        3.3.1 Curved Line Split Algorithm第70-73页
        3.3.2 Line Segmentation Algorithm第73-81页
    3.4 Ligature segmentation第81-84页
        3.4.1 Extract Information第82页
        3.4.2 Decide Primary and Secondary Components第82-83页
        3.4.3 Allocate Secondary to Primary Components第83-84页
    3.5 Results and analysis第84-95页
        3.5.1 Results of the Line Segmentation Algorithm第84-86页
        3.5.2 Comparison of Line Segmentation Algorithms第86-91页
        3.5.3 Results of the proposed ligature segmentation algorithm第91-95页
    3.6 Conclusion第95页
    3.7 Summary第95-96页
CHAPTER 4 Stacked Denoising Autoencoder based Urdu Nastaleeq LigatureRecognition第96-114页
    4.1 Introduction第96-99页
        4.1.1 Prevalent Works第96-98页
        4.1.2 Motivation第98页
        4.1.3 Contribution第98-99页
    4.2 Learning model第99-105页
        4.2.1 Autoencoder第99-100页
        4.2.2 Urdu Ligature Recognition Stacked Denoising Autoencoder (ULR-SDA)第100-105页
    4.3 Experiments第105-108页
        4.3.1 Dataset and Feature Extraction第105-106页
        4.3.2 Experiment Setup第106-107页
        4.3.3 Training ULR-SDA第107-108页
    4.4 Results第108-112页
        4.4.1 Comparison of ULR-SDA and SVM第108-109页
        4.4.2 Structure of ULR-SDA第109-110页
        4.4.3 Dimensions第110-112页
    4.5 Conclusion第112页
    4.6 Summary第112-114页
CHAPTER 5 Ligature based Urdu Nastaleeq Sentence Recognition usingGated Bidirectional Long Short Term Memory第114-135页
    5.1 Introduction第114-119页
        5.1.1 Motivation第115-119页
        5.1.2 Contribution第119页
    5.2 Overview of Recurrent Neural Network and Related Work第119-122页
    5.3 Proposed Model第122-126页
    5.4 Dataset Description and Preprocessing第126-127页
    5.5 Network Parameters第127-128页
    5.6 Experiments and Results第128-134页
        5.6.1 Comparison of GBLSTM with Prevalent LSTM based Urdu Recognition Systems第129-131页
        5.6.2 Comparison of GBLSTM with Prevalent Ligature Based Urdu RecognitionSystems第131页
        5.6.3 Number of Ligature Classes Incorrectly Predicted by GBLSTM第131-133页
        5.6.4 Top 10 Misclassified Classes第133-134页
        5.6.5 Performance Analysis on the Basis of Full Predicted Sentences第134页
    5.7 Conclusion第134页
    5.8 Summary第134-135页
Chapter 6 Conclusions and Potential Future Directions第135-138页
    6.1 Research Insights第135-137页
    6.2 Potential Future Recommendations第137-138页
Bibliography第138-151页
Acknowledgement第151-152页
List of Publications第152页

论文共152页,点击 下载论文
上一篇:基于深度学习的跨模态检索
下一篇:基于无线通信信号的目标识别关键技术研究