本文利用 K-means 擅長作分群的特性與 SLIC 超像素的加速方法作為影像的 初步分割，再搭配日亦普及的深度影像資訊來輔助，作為區塊是否合併的主要依 據。由於超像素方法處理容易造成過度分割的現象，如何將這些區塊作有效的合 併自然成為一個重要的環節。本文結合深度與色彩訊息，提出四分位數統計方法 達到物體的外框大多都可分割出來，有利於後續的影像分析與檢索。
本文方法尚有一些需要改善之處，物件的完整切割除了仍需仰賴少許的參數 外，所提方法對於深度資料的使用比重相對較高，如何獲得良好的深度資料是一 個重要的前提。在深度資料變化不明顯的情況下，合併方法便容易產生誤判。若 深度影像的空洞區域過大，排除空洞資訊的效果容易受到影響，如何改善深度影 像品質問題是未來要繼續努力的方向。其次，超像素方法會將影像分成許多的區 塊，有些區塊會被切的比較細小，原本應該要與相鄰區塊合併起來，但因區域數 值少容易造成統計出來的四分位數與相鄰區塊差異過大，可能造成錯誤合併的結 果，此一問題值得後續進一步作探討。
30 Incomplete K-means Clustering and Category Optimization,” Journal of Computer Applications, vol. 32, no. 1, pp. 248-251, 2012.
[4] 余燁、李冰飛、張小魏、劉宜萍、李海琴，面向 RGBD 圖像的標計分水嶺 分割，中國圖像圖形學報，第 21 卷，第 2 期，145-154 頁，2016 年。
[5] Mahfuzur Rahman Khan, ABM Muhitur Rahman, G.M Atiqur Rahamany and Md Abul Hasnat, “Unsupervised RGB-D Image Segmentation by Multi-layer Clustering,” International Conference on Informatics, Electronics and Vision (ICIEV), pp. 719-724, 2016.
[6] L. Cruz, D. Lucio and L. Velho, “Kinect and RGBD Images: Challenges and Applications,” IEEE Conference on Graphics, Patterns and Images, pp. 36-49, 2012.
[7] Kinect, www.cnblogs.com/TracePlus/p/4136297.html, 2014.
[8] 鍾國亮，影像處理與電腦視覺，東華書局第 5 版，2010 年。
[9] Region growing,http://blog.csdn.net/bagboy_taobao_com/article/details/5666091, 2010
[10] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua,
31
and Sabine Susstrunk, “SLIC Superpixels Compared to Stat-of-the-Art Superpixel Methods,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, no. 11, pp. 2274-2280, 2012
[11] 彩色影像與深度影像之位置對齊,
http://kheresy.wordpress.com/2011/01/21/combine_depth_and_image_from_kin ect/, 2011.
[12] Lab color space, http://zh.wikipedia.org/wiki/Lab 色彩空間, 2016 [13] K-means clustering, http://zh.wikipedia.org/wiki/K-平均算法, 2017
[14] Liangqiong Qu, Shengfeng He, Jiawei Zhang, Jiandong Tian, Yandong Tang, and Qingxiong Yang, “RGBD Salient Object Detection via Deep Fusion,”
IEEE Transactions on Image Processing, vol. 26, no. 5, pp. 2274-2284, MAY
2017.
[15] Jingyu Yang, Ziqiao Gan, Kun Li and Chunping Hou, “Graph-based Segmentation for RGB-D Data Using 3D Geometry Enhanced Superpixels,”
IEEE Transactions on Cybernetics, vol. 45, no. 5, pp. 913-926, 2015.
[16] K. Krishna and M. N. Murty, “Genetic K-means Algorithm,” IEEE Transactions on Systems, Man., and Cybernetics—Part B: Cybernetics, vol. 29, no. 4, pp.
433-439, 1999.
[17] Max Mignotte, “Segmentation by Fusion of Histogram-based K-means Clusters in Different Color Spaces,” IEEE Transactions on Image Processing, vol. 17, no.
5, pp. 780-787, 2008.
32
[18] 張桂梅、周明明、馬珂，基於彩色模型的重構標記分水嶺分割算法，中國圖 像圖形學報，第 17 卷，第 5 期，641-647 頁，2012 年。
[19] Dirk Holz, Stefan Holzer, Radu Bogdan Rusu and Sven Behnke, “Real-Time Plane Segmentation Using RGB-D Cameras,” Robot Soccer World Cup XV, LNCS 7416, pp. 306-317, 2012.
[20] Nathan Silberman, Derek Hoiem, Pushmeet Kohli and Rob Fergus, “Indoor Segmentation and Support Inference from RGBD Images,” Computer Vision – ECCV, pp. 746-760, 2012.
[21] Saurabh Gupta, Pablo Arbelaez and Jitendra Malik, “Perceptual Organization and Recognition of Indoor Scenes from RGB-D Images,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 564-571, 2013.
[22] Camille Couprie, Cl´ement Farabet, Laurent Najman and Yann LeCun, “Indoor Semantic Segmentation Using Depth Information,” First International Conference on Learning Representations (ICLR), pp. 1-8, 2013.
[23] Zhenguo Li, Xiao-Ming Wu and Shih-Fu Chang, “Segmentation Using Superpixels: A Bipartite Graph Partitioning Approach,” Computer Vision and Pattern Recognition (CVPR), pp. 789-796, 2012.