利用基因演算法於人臉辨識區別與特徵分析論文摘要

Abstract

影像辨識常常會應用到神經網路來計算影像特徵與建立模型，神經網路構造多樣化且複雜如黑盒子一般，本研究提出標準程序方法驗證神經網路的輸出結果。本研究使用Resnet模型提取128長度向量的特徵值用來辨識人臉。實驗分別使用支援向量機與基因演算法去驗證分析128向量的重要特徵。本研究最後找出其中重要特徵分配權重，提升辨識率。

Image recognition is often applied to neural networks to calculate image features and model building. Neural network structures are diverse and complex, such as black boxes. This study proposes a standard program method to verify the output of neural networks. This study uses the Resnet model to extract the eigenvalues of 128 length vectors for recognizing faces. Experiments were performed using support vector machines and gene algorithms to verify the important features of 128 vectors. At the end of the study, we found out the weight distribution of important features and improved the recognition rate.