紙鈔序號辨識

Currency Serial Number Recognition

盧毅(Yi Lu)

指導教授 ： 傅楸善

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.01447

摘要

本論文研究了一個應用導向的光學字元辨識問題：紙鈔序號辨識。一個基於統計特徵和線性分類器的方案被提出用於辨識紙鈔上的序號。在ARM9在300MHz低計算能力的數鈔機環境下，此系統可以對序號取得99.5%的準備率同時能達到每分鐘800張的處理速度。同時本文還採用三個高級的機器學習演算法：稀疏模型(Sparse Representation)，矩陣分解模型(Matrix Factorization)，梯度提升決策樹(Gradient Boosting Decision Tree)來對紙鈔序號的辨識進行研究，三個模型都能取得非常高的準確率並且具有各自的優點。在紙鈔序號辨識上的良好性能表明這三個模型對其他類型的光學字元識別也具有很大的潛力。最後論文對紙鈔字元圖像進行了二維可視化。其顯示此類數據具有嵌入在高維空間的低維隱含結構。這個結果也驗證了稀疏模型與矩陣分解模型所得到的相似結果。

關鍵字

光學字元識別 ； 紙鈔序號辨識 ； 數鈔機 ； 矩陣分解 ； 梯度提升決策樹 ； 稀疏表示 ； 數據可視化

並列摘要

We propose an application-oriented Optical Character Recognition (OCR) method for Currency Serial Number Recognition (CSNR) in this thesis. The corresponding solution based on statistical feature and linear classifier was proposed for this problem. Our proposed system could achieve the accuracy of 99.5% per bill and the speed of 800 bills per minute in the banknote counting machine with low computational power of ARM9 at 300MHz. We also apply three advanced machine learning methods including Sparse Representation (SR), Matrix Factorization (MF), Gradient Boosting Decision Tree (GBDT) for this specific OCR problem. The high recognition capacities of these methods for OCR problem are confirmed. The experiment results in CSNR have shown these methods promising candidates for more general OCR problem. The visualization of currency serial number data revealed the implicit low-dimensional structure of data that is also observed by the analytical results of MF and SR methods.

並列關鍵字

Optical Character Recognition ； Currency Serial Number Recognition ； Banknote Counting Machine ； Matrix Factorization ； Gradient Boosting Decision Tree ； Sparse Representation ； Data Visualization

參考文獻

[1] T. Ahonen, A. Hadid, and M. Pietikainen, “Face Description with Local Binary Patterns: Application to Face Recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 28, No. 12, pp. 2037-2041, 2006.

[2] E. Borovikov, “A Survey of Modern Optical Character Recognition,” http://arxiv.org/abs/1412.4183, 2015.

[3] N. Dalal and B. Triggs, “Histograms of Oriented Gradients for Human Detection,” Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Diego, CA, Vol. 1, pp. 886–893, 2005.

[4] P. Domingos, “A Unified Bias-Variance Decomposition,” Proceedings of International Conference on Machine Learning, Stanford, CA, pp. 231-238, 2000.

[5] L. Eldén, Matrix Methods in Data Mining and Pattern Recognition, Vol. 4. SIAM, Philadelphia, Pennsylvania, 2007.