比對字形拓撲結構辨認手寫字

Matching Topological Structures for Handwritten Characters Recognition

陳仰恩(Yang-En Chen)

指導教授 ： 劉長遠

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201700782

摘要

本文詳細記載指導教授所授knowhow方法，本人負責部分實作全文與學長協力完成。
論文提出了一個比對方法，用以解決在辨認手寫字時面對的各類扭曲變形。因為文字拓樸結構相當穩定，所以藉由比對此穩定的結構，容忍複雜而困難的扭曲。在抽取特徵時，文字被離散化為多個橢圓或角狀橢圓，橢圓類似生物視覺系統之接受區(Receptive fields)的形狀。文字骨架上的所有橢圓特徵的位置、角度、方向都具有一定的彈性，因此可以彈性的比對扭曲變形。對於每個樣本，首先比對出未知文字中最相似的特徵，然後平移整個樣本，使樣本的對應特徵的位置重合該未知特徵，接著就此平移後的樣本，比對出未知文字中次相似之特徵，然後以最相似特徵為中心，旋轉並縮放整個樣本，使對應於次相似未知特徵的樣本特徵重合該未知特徵，最後就此旋轉縮放後的樣本，比對整個未知文字，所有樣本特徵都比對出其最相似的未知特徵，而其相似程度的總和即為此樣本與未知文字的相似程度。

關鍵字

模式辨認 ； 影像辨識 ； 影像分類 ； 影像復原 ； 機器視覺 ； 手寫文字辨識

並列摘要

This work presents a matching process to resolve the variable distortions for recognition of handwritten characters. It can tolerate difficult distortions by preserving the topological structure of the character. The unknown character and all templates are represented in advance by features that resemble the receptive fields of visual system. The templates are loaded with flexible and variable features along their skeletons. These flexible features can tolerate local distortions. The global operations, shift, rotation, and scale, are applied to revise the whole template according to certain highly matched features. After these operations, the matched feature in the revised template will overlap its corresponding feature in the unknown character as much as possible. New matching score can be calculated for each feature of the revised template. This kind revision can be operated iteratively. The recognition is accomplished for the template which received the highest whole score. This work illustrates this matching process and its operations. This process indirectly overcomes difficult distortion problems.

並列關鍵字

pattern recognition ； image recognition ； image classification ； image restoration ； robot vision ； hand written character recognition

參考文獻

1. Blasdel, G.G. and G. Salama.: Voltage-sensitive dyes reveal a modular organization in monkey striate cortex. Nature, vol. 321, pp. 579--585 (1986)

2. Hubel, D.H. and Wiesel, T.N.: Brain Mechanisms of Vision. I. Rock, ed, The Perceptual World: Readings from Scientific American Magazine, pp. 3--24 (1990)

3. Kohonen, T., Self-organization and Associative Memory, 3rd edition. Berlin, Heidelberg, Germany: Springer-Verlag (1989)

4. Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet Classification with Deep Convolutional Neural Networks. In Advances in neural information processing systems pp. 1097--1105 (2012)

5. Liou, C. Y., Yang, H. C.: Selective Feature-to-Feature Adhesion for Recognition of Cursive Handprinted Characters. IEEE Trans. Pattern Anal. Mach. Intell., 21(2), pp. 184--191 (1999)