宋大偉,馬鳳娟,趙 華
?
基于相似度模型耦合角度制約規(guī)則的圖像匹配算法
*宋大偉1,馬鳳娟1,趙 華2
(1. 濰坊工程職業(yè)學院,山東,濰坊 262500;2. 山東科技大學,山東,青島 266590 )
圖像匹配;FAST特征檢測;SURF機制;SSIM模型;相似度模型;角度制約規(guī)則
數(shù)字圖像給人們的生活帶來了便利,為當代信息的傳遞提供了媒介[1]。人們通過數(shù)字圖像可以實現(xiàn)快速的信息傳遞以及便捷的信息儲存。目前,數(shù)字圖像匹配技術已被應用到了刑事偵查、目標追蹤以及人臉識別等多項技術范疇,是當下熱門的研究技術之一[2]。
數(shù)字圖像匹配技術的發(fā)展對人們有著重要的影響,目前出現(xiàn)了較多的數(shù)字圖像匹配方法。如Hossain等人[3]設計了利用局部信息獲取特征描述子的方法,通過局部信息的灰度等特征實現(xiàn)匹配,實驗結果表明,這種方法能夠較好地獲取圖像的特征信息,較準確地對圖像特征進行匹配。Zhao等人[4]設計了一種利用線段方法對多模態(tài)圖像進行匹配的技術,通過多模態(tài)魯棒線段描述符對圖像特征進行區(qū)分,并通過描述符的相似性檢測獲取匹配結果。張煥龍等人[5]對布谷鳥算法進行研究,將其引入圖像匹配,通過獲取圖像的HOG特征,利用布谷鳥搜索方法獲取特征匹配結果。Tsai等人[6]將不同圖像的特征描述符進行比較,設計了分類環(huán)機制,將未校正的匹配對進行濾除,提高匹配結果的正確性。
圖1 所提算法的匹配過程
對于匹配正確的特征點而言,其與特征點間構成的角度具有一定的接近度。為了進一步提高圖像特征匹配的正確率,在此,將利用特征點間角度,建立角度制約規(guī)則,對特征點進行精匹配。
3)將三對正確匹配點中的一組替換成一對新的匹配點,并返回步驟(1),若新加入的匹配點組成的角度差值滿足步驟(2)的判斷條件,則判定新加入的匹配點對為正確匹配點對,否則為錯誤匹配點對給予剔除,從而實現(xiàn)特征的精匹配。
圖2 不同算法對光照度變化圖像的匹配效果
將圖5作為圖像A,將其進行不同角度的旋轉(zhuǎn),形成圖像B。利用不同算法對圖像A與旋轉(zhuǎn)形成的圖像B進行匹配,并對匹配結果的正確度進行統(tǒng)計,以測試所提算法的匹配性能。
圖5 測試目標
圖6 匹配正確度的測試結果
[1] 湯鵬杰,譚云蘭,李金忠. 基于雙流混合變換CNN特征的圖像分類與識別[J]. 井岡山大學學報:自然科學版, 2015, 36(5): 53-59.
[2] Tony L. Image Matching Using Generalized Scale-Space Interest Points[J].Journal of Mathematical Imaging and Vision, 2015, 52(1): 3-36.
[3] Hossain M T, Shyh W T. Multimodal Image Registration Technique Based on Improved Local Feature Descriptors[J]. Journal of Electronic Imaging, 2013, 1(24): 1-17.
[4] Zhao C Y, Zhao H C, Lv J F. Multimodal Image Matching Based on Multimodality Robust Line Segment Descriptor[J]. Neurocomputing,2016,177(1):290-303.
[5] 張煥龍,張秀嬌,賀振東. 基于布谷鳥搜索的圖像匹配方法研究[J].鄭州大學學報:理學版,2017,49(4):51-56.
[6] Tsai C H, Lin Y C. An Accelerated Image Matching Technique for UAV Orthoimage Registration[J].ISPRS Journal of Photogrammetry and Remote Sensing,2017, 128(1): 130-145.
[7] 陳劍虹,韓小珍. 結合FAST-SURF和改進k-d樹最近鄰查找的圖像配準[J].西安理工大學學報,2016,32(2): 213-217.
[8] Jia D, CAO J, Song W D. Colour FAST (CFAST) Match: Fast Affine Template Matching for Colour Images [J]. Electronics Letters, 2016, 52(14): 1220-1221.
[9] 彭勃宇,王崴,周誠. 面向增強現(xiàn)實的SUSAN-SURF快速匹配算法[J]. 計算機應用研究,2015,32(8): 2538-2542.
[10] Dou J F, Qin Q, Tu Z M. Robust Image Matching with Cascaded Outliers Removal[J].Pattern Recognition and Image Analysis,2017,27(3):480-493.
[11] Beatriz O, Eva R, Jacint V. SURF-Based Mammalian Species Identification System[J]. Multimedia Tools and Applications, 2017,76(7):10133-10147.
[12] Zhang E L, Ma J, Wang X T. Improved SURF Algorithm for Color Remote Sensing Image Registration[J]. Chinese Journal of Liquid Crystals and Displays, 2017,32(2):144-152
[13] Sun Y W, Li H, Sun L. Use of Satellite Image for Constructing the Unmanned Aerial Vehicle Image Matching Framework[J].Journal of Applied Remote Sensing,2017,11(1): 1-12.
[14] Bo Y, Bahetiyaer B, Ke L. Learning Quality Assessment of Retargeted Images[J]. Signal Processing: Image Communication, 2017,56(1): 12-19.
[15] Seonyeong P, Siyong K, Byongyong Y. A Novel Method of Cone Beam CT Projection Binning Based on Image Registration[J].IEEE Trans Med Imaging,2017,36(8): 1733-1745.
[16] 吳鵬徐,洪玲,宋文龍. 結合小波金字塔的快速NCC圖像匹配算法[J].哈爾濱工程大學學報,2017,38(5): 791-796.
[17] Sebastián C C, JoséG G. SIFT Otimization and Automation for Matching Images from Multiple Temporal Sources[J]. International Journal of Applied Earth Observation and Geoinformation,2017,57(1): 113-122.
Image Matching Method Based on Similarity Model Coupling Angle Constraint Rule
*SONG Da-wei1,MA Feng-juan1,ZHAO Hua2
(1. Weifang engineering Career Academy, Weifang, Shandong 262500, China;2. Shandong University of Science and Technology, Qingdao, Shandong 266590, China)
The current image matching methods mainly achieve image matching by measuring the distance, which neglect the similarity between images and result in more mismatches and poor robustness. In this paper, an image matching algorithm based on similarity degree model and coupling angle constraint rule is proposed. High-speed and high-accuracy feature detection method is used to detect the image features, and the feature points with high accuracy can be obtained fast, which is helpful to improve the matching accuracy of the algorithm. Using the feature description mechanism, the feature points are described by calculating the wavelet response values in the feature circle domain. The structure similarity model is introduced and combined with Euclidean distance model to construct similarity model. The feature points are roughly matched from the aspects of structure similarity and measurement distance. The cosine relation of feature points is used to calculate the angle between feature points, and the angle restriction rules are established to match the feature points accurately. Experimental results show that this matching algorithm has better matching performance and higher matching accuracy compared with the typical matching method.
image matching; FAST feature detection; SURF mechanism; SSIM model; similarity model; angle constraint rule
TP391
A
10.3969/j.issn.1674-8085.2019.02.008
1674-8085(2019)02-0039-06
2018-11-23;
2018-12-27
山東省自然科學基金項目(ZR2013FQ030)
*宋大偉(1976-),男,山東濰坊人,副教授,主要從事圖像處理、計算機網(wǎng)絡技術、數(shù)據(jù)庫技術等方面的研究(E-mail: songdiv@sohu.com);
馬鳳娟(1975-),女,山東濰坊人,副教授,主要從事計算機圖像、多媒體技術、數(shù)據(jù)庫等方面的研究(E-mail: juanfm@tom.com);
趙 華(1980-),女,山東泗水人,副教授,博士,主要從事圖像處理、話題檢測與跟蹤、網(wǎng)絡輿情挖掘等方面的研究(E-mail:Zhaoh19SLK80S@163.com).