應用於全景接圖之低複雜度特徵點描述子建立與特徵點匹配演算法

Low Complexity Algorithm of Feature Point Descriptor Creation and Feature Point Matching Applied to Panoramic Image Stitching

楊家俊(Chia-Chun Yang)

指導教授 ： 曹恆偉

國立臺灣大學/電機資訊學院/電子工程學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.10775

摘要

SIFT特徵點描述子建立演算法與特徵點匹配演算法應用於全景接圖時，存在運算複雜度過高的缺點，為了改善此缺點，本論文中設計了一種應用於全景接圖的低複雜度特徵點描述子建立演算法與低複雜度特徵點匹配演算法。在特徵點描述子建立演算法中，我們使用特徵點座標間的相對關係建立特徵點描述子。在特徵點匹配演算法中，我們設計了匹配準則，並使用homography matrix 預測可能匹配的特徵點位置。
根據運算複雜度的分析與比較，我們的演算法確實比SIFT演算法的複雜度還低。另外我們使用MSSIM做客觀影像品質評估，根據客觀影像品質評估結果，使用我們的演算法可得到品質可接受的全景圖。最後我們對全景接圖演算法做定點運算模擬，根據模擬的結果，我們可以使用24位元以上的定點運算實現全景接圖，並且我們對使用24位元的定點運算所得到的全景圖進行客觀與主觀影像品質評估，根據品質評估結果，此全景圖的影像品質為可接受。
藉由使用所提出的低複雜度特徵點建立演算法與低複雜度特徵點匹配演算法應用於全景接圖，我們可達到低運算複雜度的目標，且在使用定點數運算實現時，所得到的全景圖品質可以被接受。

關鍵字

SIFT ； 全景接圖 ； 低複雜度 ； 特徵點描述子建立 ； 特徵點匹配 ； MSSIM ； 定點運算模擬

並列摘要

SIFT feature point descriptor creation algorithm and SIFT feature point matching algorithm have the drawback of high computational complexity when applied to the panoramic image stitching. To improve this drawback, we design the low complexity algorithms of feature point descriptor creation and feature point matching applicable to panoramic image stitching. In feature point matching algorithm, we use the relation between the feature point coordinates to create feature point descriptor. In feature point matching algorithm, we design the matching criteria, and use the homography matrix to predict the possibly matched feature points.
According to computational complexity analysis and comparison, the complexity of our algorithm is lower than SIFT algorithm. In addition, we use MSSIM to assess the objective image quality. According to the objective image quality assessment results, panoramic images with acceptable image quality can be obtained by using our algorithm. Finally, we do fixed-point arithmetic simulation for the panorama image stitching algorithm. According to the simulation results, we can use 24-bit fixed-point arithmetic to implement the panoramic image stitching. And we assess the objective image quality and subjective image quality of panoramic image obtained from 24-bit fixed-point arithmetic. According to the image quality assessment results, the objective image quality and subjective image quality of panoramic image are both acceptable.
By applying our low complexity feature points descriptor creation algorithm and low complexity feature point matching algorithm applied to panorama image stitching, we can achieve the goal of low computational complexity. Also, fixed-point arithmetic can be used to implement the panoramic image stitching, and the panoramic image quality has been demonstrated to be acceptable.

並列關鍵字

SIFT ； panoramic image stitching ； low complexity ； feature point descriptor creation ； feature point matching ； MSSIM ； fixed-point arithmetic simulation

參考文獻

[20] 孫維辰，〝結合顏色與空間特性的物件追蹤演算法〞，國立台北科技大學資訊工程系碩士論文，2009年7月。

[2] E. D. Castro, C. Morandi, “Registration of translated and rotated images using finite Fourier transforms,” IEEE Transaction on Pattern Analysis and Machine Intelligence, 9(5):700-703, 1987.

[3] B. Reddy, B.Chatterji, “An FFT-based technique for translation, rotation, and scale-invariant image registration,” IEEE Transactions on Image Processing, 5:1266-1271, 1996.

[4] Richard Szeliski, “Video mosaics for virtual environments,” IEEE Computer Graphics and Applications, 16(2):22-31, 1996.

[6] Matthew Brown and David G. Lowe, “Automatic panoramic image stitching using invariant features”, International Journal of Computer Vision, 74(1):59-73, 2007.