由於網際網路上的影像資料日益增多，為使影像資料建立分類目錄時，能有效進行影像自動分類，本論文提出採用影像循序特徵建立決策樹以產生分類規則的影像自動分類方法。首先在影像特徵擷取部份，將影像的顏色從RGB色空間轉換至HSI色空間，經由色彩量化取出色塊當作影像物件，以色塊不同顏色值做為影像特徵項，再依其色塊的位置屬性值大小排序產生影像特徵項序列。接下來利用循序探勘法找出所有常見特徵項序列、最大常見特徵項序列、及最長常見特徵項序列，將這些特徵項序列定為影像的分類屬性，並依影像是否包含特徵項序列來設定該屬性值為0或1，最後套用C4.5建立決策樹並產生影像分類規則。為了讓系統能漸進式學習，當系統分類結果由人為判斷是錯誤時，分類錯誤影像的特徵可再加入訓練樣本，以期增進系統的分類正確率。為了在重新探勘分類規則時較有效率，系統除保留原訓練資料集的常見特徵項序列的資訊外並保留負邊界樣式，以漸進式探勘方式找出新分類屬性，以節省再次重新掃描所有影像特徵項序列的時間。實驗評估結果顯示，本論文所提出之影像分類方法於不同類型之影像皆可達到不錯的分類效果。此外，比較另一個影像分類方法，本方法達到較好的分類正確率，且在分類規則比對時需較少的比較次數。

In this thesis, a method of image classifications is proposed. This approach is designed based on constructing decision trees for sequential patterns. First, the color space of images is transferred from RGB to HSI. After performing quantization on the color space, color blocks in an image are extracted and blocks with the same color are assigned the same identifiers of feature terms. According to y-positions of color blocks, blocks are sorted to form a sequence of feature terms in order to represent features of an image. Frequent sequential patterns, mined from the sequences of image feature terms extracted from training images, are used to be the attributes for classification. Finally, according to the selected attributes, a decision tree is constructed by performing C4.5 algorithm to find the classification rules. Moreover, in order to improve the accurate rate of classification, new images which are assigned the wrong categories by the system can be inserted into training set to re-train the classification rules. For achieving more efficient performance when performing re-training, the concept of incremental mining is applied in the system to preserve the information of frequent sequential patterns and negative borders in the previous training images. Such that it prevents re-scanning the whole training data set to select the new classification attributes. The experiment results show that the accurate rates of the proposed method is good for various kinds of image. Furthermore, by comparing with another related work, our method has better accurate rate and has less numbers of comparisons when searching classification rules.

[1] A. Vailaya, A. Jain, M. Figueiredo, and H. Zhang, “Content-Based Hierarchical Classification of Vacation Images, ” IEEE International Conference on Multimedia Computing and Systems Volume I-Volume 1, 1999.
[2] C.H. Lo and S.Y. Chen,” General Image Classification Using Adaptive Cellular Color Decomposition,” International Journal of Pattern Recognition and Artificial Intelligence, accepted and to appear, 2003. (SCI, EI)
[3] G. Becker and P. Bock, “Shape Classification Using a Radial Feature Token,” 29th Applied Imagery Pattern Recognition Workshop (AIPR'00), 2000.
[4] J. Pei, J. Han, B. Mortazavi-Asl, H. Pinto, Q. Chen, U. Dayal, and M. –C. Hsu, “PrefixSpan: Mining Sequential Patterns Efficiently by Prefix-Projected Pattern Growth,” in proceeding of IEEE int. Conf. On Data Engineering(ICD’01), 2001.
[5] K. Hirata, S. Mukherjea, W.S. Li, and Y. Hara, “Integrating Image Matching and Classification for Multimedia Retrieval on the Web, ” IEEE International Conference on Multimedia Computing and Systems Volume I-Volume 1, 1999.
[6] L. Breiman, J. H. Friedman, R.A. Olshen, and C.J. Stone, ”Classification and Regression Trees,” Belmont, CA: Wadsworth International Group, 1984.
[7] M.J.Zaki, “Efficient Enumeration of Frequent Sequences,” In 7th CIKM, 1998.
[8] N. Vandenbroucke, L. Macaire, and J.G. Postaire, “Color Image Segmentation by Supervised Pixel Classification in a Color Texture Feature Space: Application to Soccer Image Segmentation,” International Conference on Pattern Recognition (ICPR'00)-Volume 3 , 2000.
[9] P. Scheunders, “A comparison of clustering algorithms applied to color image quantization,” Pattern Recognition Letters, vol. 18, pp. 1379-1384, 1997.
[10] R. Agrawal and R. Srikant, “Mining Sequential Pattern,” Proceedings of the 11th International Conference on Data Enginerring, Taipei, Taiwan, 1995, pp3-14.
[11] R. Srikant and R. Agrawal, “Mining Sequential Patterns: Generalizations and Performance Improvements,” Proceedings of the 5th International Conference on Extending Database Technology, Avignon, France, 1996, pp.3-17.
[12] S.D. Bona and O. Salvetti,”An Enhanced Neural System for Biomedical Image Classification,” In Proc. of 4th IEEE Southwest Symposium on Image Analysis and Interpretation, pp. 141-145, IEEE Computer Society, Austin (Texas), 2 - 4 April, 2000.
[13] S. Parthasarathy, M. J. Zaki, M. Ogihara, and S. Dwarkadas, “Incremental and Interactive Sequence Mining,” In Proceedings of the 8th International conference on Information and Knowledge Management (CIKM’99), Kansas City, MO, USA, November 1999, pp.251-258.
[14] S. Z. Li, K. L. Chan, and C. Wang, “Performance Evaluation of the Nearest Feature Line Method in Image Classification and Retrieval,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 22, No. 11, November 2000.
[15] T. Ko, and P. Bock, “Image-Content Classification Using a Dynamically Allocated ALISA Texture Module,” 29th Applied Imagery Pattern Recognition Workshop (AIPR'00), October 16 - 18, 2000.
[16] X. Wan and C.C. J. Kuo, “A new approach to image retrieval with hierarchical color clustering,” IEEE Trans. Circuits and Systems for Video Technology, Vol. 8, No. 5, 1998.
[17] 安寶楹, 柯佳伶, “以循序特徵關聯方式探勘影像分類規則之研究,” 2002.
[18] 陳建宏, 柯佳伶, “以色塊屬性關聯規則建立影像分類決策,” in Proceedings of 2001 National Computer Symposium, 2001.