基於PCA與LDA之模糊類神經網路於影像分類

PCA and LDA Based Fuzzy-Neural-Networks for Image Classification

李安泰(An-Tai Li)

指導教授 ： 吳俊德

國立暨南國際大學/科技學院/電機工程學系/碩士(2013年)

摘要

本篇論文將模糊類神經網路應用於影像分類。影像部分，利用Principal Component Analysis (PCA)、Linear Discriminant Analysis (LDA)將影像進行處理。PCA找出投影向量，使投影後的資料有最小的變異度，因此可用此來壓縮資料量且能保留大部分的原始特性。LDA則找出投影向量，使投影後的資料，群與群間的中心距離遠離，而群內的變異度減少，使資料的鑑別度增加。但是PCA部分的運算量很大，在此加入了影像處理中的Histogram的概念，把圖片中每個像素的灰階值或GRB值進行統計，可減少PCA維度的計算。

模糊類神經網路在此作為分類器，影像經過PCA與LDA的計算後的資料，作為模糊類神經網路的輸入進行分類，在此模糊化的部分採用高斯模型，參數學習採用梯度下降法來調整高斯模型。在此結合之前學長韌體設計的部分，利用實驗室目前最新版本的CPU指令，用組合語言實現模糊類神經網路中testing的部分，在高斯計算的指數部分，採用查表法，比起泰勒展開式可減少計算量，增加程式運行的效率。

關鍵字

影像分類 ； 模糊類神經網路 ； PCA ； LDA ； Histogram ； 韌體設計

並列摘要

The Fuzzy Neural Network (FNN) is applied for Image Classification in this thesis. The image data is processed by Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). PCA maximizes the variance of the samples in the projection subspace. Consequently, the image data can possesses most of the original characteristic after compression by PCA mapping. LDA uses image class information to find a subspace for better discrimination of different face classes. To reduce the dimension calculation of PCA, the Histogram method is proposed to count the RGB or gray level value of each pixel.

The FNN is a classifier in this thesis. The image data after PCA and LDA mapping is the input of the FNN. In the Fuzzification part, we use the Gaussian mixture model (GMM). Parameter learning adopts the gradient method to adjust the shapes of Gaussian functions. The testing part of the FNN is implemented by firmware coded in assembly. The used CPU is designed by our lab. To enhance the efficiency of the program, we develop some instruction for math calculation. In the exponential part, the lookup table is adopted. In contrast to Taylor expansion, the calculation is reduced obviously by the lookup table.

並列關鍵字

image classification ； FNN ； PCA ； LDA ； Histogram ； firmware

參考文獻

[1] M. Kirby and L. Sirovich, “Application of the karhunen-loeve procedure for the characterization of human faces,” IEEE Trans. on Pattern Analysis and Machine Intelligence, 12(1), January 1990.

[2] Turk, M. and A. Pentland. “Eigenfaces for recognition,” Journal of Cognitive Neuroscience, 1991. 3(1): p. 71-86.

[3] P.N. Belhumeur, J.P. Hespanha, and D.J. Kriegman, “Eigenfaces vs. Fisherfaces: Recognition using class specific linear projection,” IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp.711-720, July 1997.

[4] W. Zhao, R. Chellapa, A. Krishnaswamy, “Discriminant analysis of principal components for face recognition,” In Proceedings of the 3rd International Conference on Automatic Face and Gesture Recognition, pages 336-341, 1998.

[5] D. Swets and J. Weng, “Using discriminant eigenfeatures for image retrieval,” IEEE Trans. on Pattern Analysis and Machine Intelligence, 18:831-836, 1996.