臺灣博碩士論文加值系統

科學家為了仿效靈長類動物視覺，藉由參考視覺皮層中的腹側流，發明了一種階層式模型HMAX (Hierarchical Model and X)，其中，HMAX 的有效性取決於其「學習樣本」的質量。然而，由於HMAX 樣本採取隨機方式選擇，導致其學習機制的不穩定；為了解決這個問題，本研究提出一種基於顯著圖和顯著區域的新策略。該方法在選擇學習樣本以前使用顯著性圖先分析C1 特徵。爾後，設置門檻值區分顯著區域，用以辨別隨機選取之樣本質量。為了證明策略效果，使用Caltech101 資料庫進行多類別實驗分類，並與原始HMAX 相比較。實驗結果顯示，該方法可以減少學習樣本數量，同時也僅略為影響準確度。

HMAX, which is called Hierarchical Model and X, is a hierarchical model invented by scientists to imitate the ventral stream in the visual cortex of primates. The performance of HMAX depends heavily on the quality of its learning patches. However, HMAX’s learning mechanism is unstable due to those learning patches but are selected randomly. In order to address the problem, a strategy by leveraging saliency map and salient region is proposed in this paper. During the feature learning stage, the proposed method analyzes the feature maps of C1 layer to generate saliency maps before learning patches selection. After that, a threshold, which is used to distinguish the quality of the candidates of learning patches, is set to separate the salient regions from the saliency maps. To evaluate the performance of the proposed strategy, several experiments on multi-category classification problems using the Caltech101 database are conducted to compete with original HMAX. The experiment results show that the proposed method can decrease the amount of learning patches, but also slightly reduce the classification accuracy.

目次
中文摘要 i
Abstract ii
致謝 iii
目次 iv
表次 vi
圖次 vii
第一章 緒論 1
1.1 研究背景 1
1.2 研究目標 2
1.3 論文架構 2
第二章 相關研究 4
2.1 視覺與大腦 4
2.2 HMAX 8
2.2.1 原始的HMAX架構 8
2.2.2 基於圖像金字塔之HMAX架構 14
第三章 研究架構 16
3.1 顯著圖與顯著區域 16
3.2 學習樣本挑選策略 18
3.3 策略參數 22
3.3.1 顯著圖計算方法 22
3.3.2 顯著區域門檻 24
3.3.3 顯著樣本門檻 24
第四章 實驗結果 25
4.1 實驗環境 25
4.2 多類別分類實驗 25
4.2.1 綜合結果 27
4.2.2 下降比率 29
4.2.3 顯著區域門檻比較 30
4.2.4 顯著圖計算方式比較 30
4.2.5 顯著區域門檻倍率比較 31
4.2.6 顯著樣本門檻比例比較 31
4.3 實驗結論 31
第五章 結論與未來展望 35
5.1 結論 35
5.2 未來研究方向 35
參考資料 36

[1] D. H. Hubel and T. N. Wiesel, "Receptive fields, binocular interaction and functional architecture in the cat's visual cortex," The Journal of Physiology, vol. 160, pp. 106-154, 1962.
[2] D. H. Hubel and T. N. Wiesel, "Receptive fields and functional architecture in two nonstriate visual areas (18 and 19) of the cat," Journal of Neurophysiology, vol. 28, no. 2, pp. 229-289, 1965.
[3] M. A. Goodale and D. A. Milner, "Separate visual pathways for perception and action," Trends in Neurosciences, vol. 15, no. 1, pp. 20-25, 1992.
[4] H. Gray, Anatomy of the Human Body, vol. 8, Lea & Febiger, 1878.
[5] M. Mishkin and U. Leslie G., "Contribution of striate inputs to the visuospatial functions of parieto-preoccipital cortex in monkeys," Behavioural Brain Research, vol. 6, no. 1, pp. 57-77, 1982.
[6] L. G. Ungerleider and J. V. Haxby, "'What'and 'where'in the human brain," Current Opinion in Neurobiology, vol. 4, no. 2, pp. 157-165, 1994.
[7] R. Snowden, P. Thompson, and T. Troscianko, Basic vision: an introduction to visual perception, Oxford University Press, 2012, pp. 81, 83.
[8] M. Riesenhuber and T. Poggio, "Hierarchical models of object recognition in cortex," Nature Neuroscience, vol. 2, pp. 1019-1025, 1999.
[9] S. Marĉelja, "Mathematical description of the responses of simple cortical cells," Journal of the Optical Society of America, vol. 70, no. 11, pp. 1297-1300, 1980.
[10] T. Serre and M. Riesenhuber, "Realistic modeling of simple and complex cell tuning in the HMAX model, and implications for invariant object recognition in cortex," CBCL Paper 239 / AI Memo 2004-017, Massachusetts Institute of Technology, Cambridge, MA, 2004.
[11] J. Mutch and D. G. Lowe, "Object class recognition and localization using sparse features with limited receptive fields," International Journal of Computer Vision, vol. 80, no. 1, pp. 45-57, 2008.
[12] S. Pannasch, J. R. Helmert, K. Roth, A. K. Herbold, and H. Walter, "Visual fixation durations and saccade amplitudes: shifting relationship in a variety of conditions," Journal of Eye Movement Research, vol. 2, pp. 1-19, 2002.
[13] W. van Zoest and M. Donk, "Bottom-up and top-down control in visual search," Perception, vol. 33, no. 8, pp. 927-937, 2004.
[14] L. Itti, C. Koch and E. Niebur, "A model of saliency-based visual attention for rapid scene analysis," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, no. 11, pp. 1254-1259, 1998.
[15] X. Hou and L. Zhang, "Saliency detection: A spectral residual approach," in Proc. of IEEE Conference on Computer Vision and Pattern Recognition, pp. 1-8, 2007.
[16] Y. F. Lu, T. K. Kang, H. Z. Zhang, and M. T. Lim, "Enhanced hierarchical model of object recognition based on a novel patch selection method in salient regions," IET Computer Vision, vol. 9, no. 5, pp. 663-672, 2015.
[17] R. Rifkin, G. Yeo and T. Poggio, "Regularized least-squares classification," Nato Science Series Sub Series III Computer and Systems Sciences, vol. 190, pp. 131-154, 2003.
[18] F. F. Li, F. Rob and P. Pietro, "Learning generative visual models from few training examples: an incremental bayesian approach tested on 101 object categories," Computer Vision and Image Understanding, vol. 106, no. 1, pp. 59-70, 2007.
[19] J. Mutch, U. Knoblich, and T. Poggio, "CNS: a GPU-based framework for simulating cortically-organized networks," MIT-CSAIL-TR-2010-013 / CBCL-286, Cambridge, MA, 2010.