手寫數字辨識模型之比較研究

A Comparison of Handwritten Digits Recognition Models

本研究主要目的為比較兩種不同的演算法對手寫數字識別之表現。本研究採用支援向量機（SVM）及類神經網路（ANN）作為比較的項目，並且在資料集的處理上加以上主成分分析（PCA）與OpenCV為基礎的資料預處理。本實驗基於支援向量機、類神經網路、主成分分析以及OpenCV建構了一個手寫數字辨識模型（Hand-written Digits Recognition Model），並用此模型所訓練出來的演算法分類器設計一系統來實測實際手寫的數字影像。實驗的結果為使用支援向量機的準確率優於類神經網路，而在使用了主成分分析將資料集進行預處理降維後，在準確率不變的情況下，大幅的減少了其運算的時間。 The purpose of our study is to compare the performance in two different kinds of machine learning algorithms on handwritten digits recognition. Our study choose support vector machine (SVM) and artificial neural network (ANN) as the comparing algorithms, and use OpenCV and principal components analysis (PCA) to preprocess the dataset. The result shows that SVM has the better performance than ANN, and the dataset preprocessing based on PCA can improve the efficiency when calculating the result. Moreover we construct a HDI model based on SVM, ANN, PCA and OpenCV, and use the classifier training from this HDI model to build a system that can recognize the handwritten digits written by ourselves.