基於紋理特徵之花卉辨識

Flower Recognition Based on Textural Feature

黃瑞賢(Rui-Xian Huang)

指導教授 ： 蘇義明

共同指導教授 ： 林文寬(Wen-Kuan Lin)

義守大學/電機資訊學院/電子工程學系/碩士(2015年)

https://doi.org/10.6343/ISU.2015.00256

摘要

本論文目的是在實現一個花卉分類系統，該系統包括Grab-cut花卉分割，紋理(textural)特徵抽取與K鄰近分類(KNN)等多個處理步驟組合而成。首先花卉分割步驟採用圖片內縮50像素的範圍定為目標可能區域，接下來是分割的估計，用範圍外的像素參考為背景找出區域內的目標像素，並執行最小切割，接著再進行目標區域的估計，重複執行分割與估計直到圖像收斂接著得到一個完整的花卉主體。接著使用紋理特徵關鍵點抽取步驟，而紋理特徵是由斑狀結構、連接點、和邊緣資訊所組成，為了抽取這些關鍵點，此步驟採用加速穩健特徵抽取(SURF)方法，此方法將花卉的主體進行積分影像，並用荷希(Hessian)矩陣濾波找出花卉的特徵關鍵點，並用哈爾(Haar)小波中的水平方向與垂直方向找出關鍵點的矢量方向，計算出一個64維的特徵方向向量，接著再把這些特徵方向向量輸入到K鄰近分類(KNN)裡進行分類，經實驗證實用於常見的花卉，使用加速穩健特徵方向向量描述來進行花卉分類系統正確率可以達到70%以上。

關鍵字

Grab-cut 花卉分割 ； 加速穩健特徵(SURF)抽取 ； KNN分類

並列摘要

The purpose of this thesis is to achieve a flower classification system, which includes a Grab-cut flowers segmentation, textural feature extraction and k-Nearest Neighbors classification (KNN) approaches. Firstly, flower segmentation step uses an image shrinking the range of 50 pixels with origninal flower image as the target area. The next segmation estimatation processing is used with a range of reference as background pixels outside the target pixel region to find the target pixels. The estimated target area with a flower object is repeatly selected with mininus cutting until the flower image is found. Some key points of the flower image are then extracted in the texture feature extraction step, but the texture is characterized by a porphyritic structure, connection points, and the edge information. In order to extract these key points, the speed up robust feature (SURF) extraction approach uses an integral image and uses Hessian matrix filtering feature to find the key points of flowers. The vertical and horizontal direction of key points are found with Haar wavelet into calculate a 64-dimensional feature direction vector. The KNN classification approach is used to recognize a flower image by the directional features. The experimental results show the flower classification system can archive about 70% of the correct rate.

並列關鍵字

Flowers segmentation ； Flower classification ； SURF feature extraction ； Grab-cut ； k-Nearest Neighbors algorithm

參考文獻

[1] Bookstein, F. L. (1996). Biometrics, biomathematics and the morphometric synthesis. Bulletin of mathematical biology, 58(2), 313-365.

[2] Kittler, J., Li, Y. P., Matas, J., & Sánchez, M. R. (1997, January). Combining evidence in multimodal personal identity recognition systems. In Audio-and Video-based Biometric Person Authentication (pp. 327-334). Springer Berlin Heidelberg.

[3] Starner, T., Auxier, J., Ashbrook, D., & Gandy, M. (2000, October). The gesture pendant: A self-illuminating, wearable, infrared computer vision system for home automation control and medical monitoring. In Wearable computers, the fourth international symposium on (pp. 87-94).

[4] Rother, C., Kolmogorov, V., & Blake, A. (2004, August). Grabcut: Interactive foreground extraction using iterated graph cuts. In ACM Transactions on Graphics (TOG) (Vol. 23, No. 3, pp. 309-314). ACM. Bay, Herbert, et al. "Speeded-up robust features (SURF)." Computer vision and image understanding 110.3 (2008): 346-359.

[5] Bay, H., Ess, A., Tuytelaars, T., & Van Gool, L. (2008). Speeded-up robust features (SURF). Computer vision and image understanding, 110(3), 346-359.