方智文
【摘 要】本文提出了一種基于共線梯度特征(collinear gradient-enhanced coding, CGEC)和結(jié)構(gòu)化輸出的異譜圖像匹配算法。首先通過(guò)共線梯度特征對(duì)異譜圖像進(jìn)行特征圖像轉(zhuǎn)換,獲得具有相似結(jié)構(gòu)信息的特征圖;接著,在特征圖上進(jìn)行關(guān)鍵點(diǎn)的提取,以得到較多的同名點(diǎn);最后,綜合關(guān)鍵點(diǎn)的特征向量和關(guān)鍵點(diǎn)之間的結(jié)構(gòu)信息,采用優(yōu)化的方法計(jì)算得到最優(yōu)的轉(zhuǎn)換關(guān)系。實(shí)驗(yàn)結(jié)果表明,本文方法更好的獲取了關(guān)鍵點(diǎn)的匹配結(jié)果。
【關(guān)鍵詞】異譜圖像匹配;共線梯度特征;結(jié)構(gòu)化輸出;關(guān)鍵點(diǎn)
【Abstract】A multi-spectral image matching algorithm based on collinear gradient-enhanced coding (CGEC) and structure output is proposed. First, the similar structure feature maps are obtained from the multi-spectral images through collinear gradient-enhanced coding. Secondly, the key points are extracted on the structure feature maps in order to get more valid point-pairs. Lastly, combining with the feature vector and structure information, we use the optimization algorithm to obtain the best transformation. The experiment result demonstrates that our method achieves better result.
【Key words】Multi-spectral image matching; Collinear gradient-enhanced coding(CGEC); Structure output; Key points
圖像匹配是機(jī)器視覺(jué)領(lǐng)域的關(guān)鍵技術(shù)之一,被廣泛的用于醫(yī)學(xué)圖像、遙感圖像、機(jī)器人視覺(jué)、導(dǎo)航等領(lǐng)域[1-3]。異譜圖像匹配是指針對(duì)不同譜段下獲取的兩張或多張圖像,找到圖像間的空間變換,建立圖像間的對(duì)應(yīng)關(guān)系。異譜圖像因其局部特征的顯著差異,大大增加了異譜圖像之間的匹配難度。傳統(tǒng)的點(diǎn)匹配方法,如尺度不變特征變換描述子(scale-invariant, SIFT)[4],高速魯棒描述子(speeded up robust features, SURF)[5],二值魯棒獨(dú)立元素描述子(binary robust independent elementary features, BRIEF)[6]等,均不能很好解決異譜圖像的局部特征差異大的問(wèn)題。本文采用共線梯度特征和結(jié)構(gòu)化輸出的方法實(shí)現(xiàn)異譜圖像對(duì)的匹配。
1 生成共線梯度特征圖
文獻(xiàn)[7]提取了共線梯度特征對(duì)異譜圖像的結(jié)構(gòu)信息進(jìn)行描述,有效的提取了異譜圖像之間的相近結(jié)構(gòu)信息。
2 關(guān)鍵點(diǎn)提取
文獻(xiàn)[7]在原始異譜圖像上直接提取FAST關(guān)鍵點(diǎn),再進(jìn)行同名點(diǎn)的匹配。但因異譜圖像在局部特性上的較大差別,導(dǎo)致關(guān)鍵點(diǎn)的位置差別較大,如可見(jiàn)光圖像具有豐富的圖像細(xì)節(jié),可提取大量的關(guān)鍵點(diǎn),但紅外圖像因熱成像的原理,往往使得多數(shù)局部區(qū)域細(xì)節(jié)消失,提出的關(guān)鍵點(diǎn)數(shù)量非常少。為了解決異譜圖像上提取的同名點(diǎn)對(duì)較少的問(wèn)題,本文采用在共線梯度特征圖上提取FAST關(guān)鍵點(diǎn)的方法。因共線梯度特征圖已將異譜圖像投影到相似的結(jié)構(gòu)信息上,因此比直接從原始圖像上能提取更多的同名點(diǎn)對(duì),有利于異譜圖像變換矩陣的正確計(jì)算。
3 結(jié)構(gòu)化輸出
4 實(shí)驗(yàn)結(jié)果
本文對(duì)紅外圖像和可見(jiàn)光圖像的異譜對(duì)進(jìn)行了點(diǎn)匹配實(shí)驗(yàn),如圖1所示。圖1左則為可見(jiàn)光圖像,右側(cè)為紅外圖像,第一行為文獻(xiàn)[7]的點(diǎn)匹配結(jié)果圖,第二行為本文的匹配結(jié)果圖。從圖1可以看出,本文方法大大增加了同名點(diǎn)的匹配數(shù)目。
第一行為文獻(xiàn)[7]的結(jié)果,第二行為本文的結(jié)果.
5 結(jié)論
本文方法基于共線梯度特征生成異譜圖像的特征圖,得到異譜圖像的大結(jié)構(gòu)信息。改善了從原始圖像上提取同名點(diǎn)對(duì)少的問(wèn)題,并通對(duì)點(diǎn)與點(diǎn)之間的結(jié)構(gòu)化信息獲取最優(yōu)化的點(diǎn)對(duì)匹配結(jié)果,大大提升了異譜圖像的點(diǎn)匹配效果。
【參考文獻(xiàn)】
[1]余時(shí)偉,黃廷祝,劉曉云,陳武凡.顯著圖引導(dǎo)下基于偏互信息的醫(yī)學(xué)圖像配準(zhǔn)[J].儀器儀表學(xué)報(bào),2013,34(6):1218-1224.
[2]葛盼盼,陳強(qiáng),顧一禾.基于Harris角點(diǎn)和SURF特征的遙感圖像匹配算法[J].計(jì)算機(jī)應(yīng)用研究,2014,31(7):2205-2208.
[3]謝文吉,孫曉剛,張亮.基于尺度不變特征變換特征點(diǎn)應(yīng)用于印刷檢測(cè)的快速匹配算法[J].計(jì)算機(jī)應(yīng)用,2013,3(S1):186-189.
[4]D. G. Lowe. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110.
[5]H. Bay, A. Ess, T. Tuytelaars, et al. Speeded-up robust features(SURF)[J]. Computer Vision and Image Understanding, 2008, 110(3): 346-359.
[6]M. Calonder, V. Lepetit, C. Strecha, et al. BRIEF: binary robust independent elementary features. Proc. of the European Conference on Computer Vision, 2010: 778-792.
[7]Qin Y, Cao Z, Zhuo W, et al. Robust key point descriptor for multi-spectral image matching[J]. Journal of Systems Engineering and Electronics, 2014, 25(4): 681.
[8]Hare S, Saffari A, Torr P H S. Efficient online structured output learning for keypoint-based object tracking[C]//Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on. IEEE, 2012: 1894-1901.
[9]Fischler, M.A. and Bolles, R.C. Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography[J]. Communications of the ACM, 1981, 4(6): 381-395.
[責(zé)任編輯:楊玉潔]