無參數需求之高精準度與高強健性影像切割演算法

High Accuracy and High Robust Natural Image Segmentation Algorithm without Parameter Adjusting

盧奕帆(I-Fan Lu)

指導教授 ： 丁建均

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.00582

摘要

在電腦視覺和影像處理的領域中，影像分割一直是重要的基礎工作。雖然本主題已經被研究了許多年，然而，要如何在全自動、無需調整參數的前提下，仍然可以將大部分的自然影像精準的分割，仍然是一項具有挑戰性的任務。此外，近年來關於超像素(superpixel) 的研究有很大的進展，這種新技術使傳統的影像分割演算法具有更高的效率和更好的性能。在這篇論文中，我們將提出一種運用超像素等多項技術的「全自動」影像分割方法，使用者無需輸入參數或是調整參數即可得到高精確度的影像分割結果。
我們的演算法採用了基於熵率的超像素(ERSs)、邊緣偵測、顯著影像偵測以及計算紋理特徵等技術。將原始影像轉為基於熵率的超像素表示，使得演算法效率大幅提高。透過計算超像素周邊以及內部的梯度資訊，傳統的邊緣偵測得以修正，進一步防止超像素過度合併。利用顯著影像偵測與計算紋理特徵，超像素合併的門檻值可根據影像作自適應調整，避免影像過度分割。模擬結果顯示，對於任意自然影像的分割處理，我們方法表現的分割結果相當符合人類感知，且不需要額外的參數調整，而這也超越了其他現有已知最先進的方法的模擬結果。

關鍵字

影像分割 ； 基於熵率的超像素 ； 邊緣偵測 ； 顯著影像偵測 ； 電腦視覺

並列摘要

In computer vision and image processing, image segmentation is always an important fundamental work. Though this topic has been researched for many years, it is still a challenging task to well segment most of the natural images automatically without adjusting any parameter. Recently, the researches of superpixels have great improvement. This new technique makes the traditional segmentation algorithms more efficient and has better performances. In this thesis, an automatic image segmentation algorithm based on superpixels and many other techniques is proposed. It can accurately segment almost all of the natural images without parameter adjustment.
In our algorithm, the techniques of entropy rate superpixels (ERSs), edge detection, saliency detection, and computing texture feature are adopted. With the aid of ERSs, the proposed algorithm can be implemented very efficiently. To prevent over-merge of superpixels, modified edge detection which computes the gradient information of the contours and the interiors of superpixels is used. Saliency detection and the texture features of an image are also used to prevent over-segmentation. Moreover, an adaptive threshold is also used for superpixel merging. These techniques make the segmentation result more consistent with human perception without adjusting any parameter. Simulations show that our proposed method can well segment most of natural images and outperform state-of-the-art methods.

並列關鍵字

Image segmentation ； ERS superpixels ； edge detection ； saliency detection ； computer vision

參考文獻

A. Image Segmentation

[1] J. Shi and J. Malik, “Normalized Cuts and Image Segmentation,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 22, no. 8, pp. 888-905, Aug. 2000.

[2] P. Felzenszwalb and D. Huttenlocher, “Efficient Graph-Based Image Segmentation,” Int’l J. Computer Vision, vol. 59, no. 2, pp. 167-181, Sept. 2004.

[4] D. Comaniciu and P. Meer, “Mean Shift: A Robust Approach toward Feature Space Analysis,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 24, no. 5, pp. 603-619, May 2002.

[5] A. Vedaldi and S. Soatto, “Quick Shift and Kernel Methods for Mode Seeking,” Proc. European Conf. Computer Vision, 2008.