張 昊
?
基于圓域梯度信息耦合角度相似法則的圖像匹配算法
張 昊
(安徽城市管理職業(yè)學(xué)院信息技術(shù)學(xué)院,安徽,合肥 230011)
為了提高算法匹配正確度與抗仿射變換性,本文通過將圖像的梯度信息引入圖像匹配的過程,提出了基于圓域梯度信息耦合角度相似法則的圖像匹配算法。首先,采用Forstner算子精準(zhǔn)快速地對圖像特征進行檢測,獲取純度較高的特征點。然后,通過計算圓域的小波響應(yīng)值確定主方向,并利用圓域的梯度信息獲取特征向量,完成特征描述。最后,根據(jù)匹配特征的仿射不變性,利用主方向建立角度相似法則,以完成匹配。通過特征點的特征向量去除錯誤匹配點,獲取最終的匹配結(jié)果。實驗數(shù)據(jù)顯示,所提算法具有較高的匹配正確率與魯棒性,在多種幾何變換下,其匹配結(jié)果中具有較多的正確匹配對,而錯誤匹配對數(shù)量較少。
圖像匹配;Forstner算子;圓域梯度信息;Haar小波;角度相似法則
計算機技術(shù)的高速發(fā)展為圖像處理技術(shù)創(chuàng)造了良好的提升環(huán)境,使得圖像處理技術(shù)得以廣泛地使用。圖像匹配已被人們用于目標(biāo)追蹤、模式識別以及安防保衛(wèi)等多個領(lǐng)域[1-2]。
所提的基于圓域梯度信息耦合角度相似法則的圖像匹配算法過程見圖1。從圖1可知,所提算法分為特征檢測、特征描述、特征匹配三個部分組成。在特征檢測階段,通過Forstner算子對精確快速地檢測圖像特征。在特征描述時,利用Haar小波響應(yīng)值獲取主方向,通過圓域梯度信息得到低維特征向量,降低算法的耗時。在特征匹配階段,算法通過特征點的主方向建立角度相似法則,對特征點進行匹配。通過特征向量去除誤匹配點,實現(xiàn)對匹配點的優(yōu)化。
圖1 所提圖像匹配算法的過程
式中,p代表的灰度值。
圖2 求取Robert梯度的示意圖
式中,表示的行列式,分別的跡。
其中,g為第維上的梯度值。
圖3 特征描述過程的示意圖
Fig.3 Schematic diagram of feature description process
通過距離測量的方式來完成特征匹配是一種較為常見的方法,該方法雖然簡單、快速,但對于放射變化魯棒性較差,存在錯誤匹配現(xiàn)象[13]。當(dāng)圖像發(fā)生旋轉(zhuǎn)等仿射變換時,特征點之間的距離就會隨之改變,而其對應(yīng)的主方向角度的差值,與旋轉(zhuǎn)角度的比值還是保持著相似性[14-15]。因此,本文將通過特征點的主方向建立角度相似法則,對特征點進行匹配,以提高算法的匹配正確度以及魯棒性。
對于一對匹配特征點,其特征向量中的元素應(yīng)該具有高度的近似性,偏差度較小[16-17]。為進一步確定匹配特征點的正確性,將錯誤匹配點進行去除。本文通過特征向量,構(gòu)建優(yōu)化函數(shù),對匹配正確性進行判斷,實現(xiàn)匹配結(jié)果優(yōu)化。
圖4 不同算法的匹配結(jié)果圖
表1 圖4中不同算法的匹配統(tǒng)計結(jié)果
表2 圖5中不同算法的匹配統(tǒng)計結(jié)果
表3 圖6中不同算法的匹配統(tǒng)計結(jié)果
表4 不同算法的平均耗時
圖7 匹配正確度測試
[1] Qi B J, Liu J G, Zhang B Y. Research on matching performance of SIFT and SURF algorithms for high resolution remote sensing image[J]. Chinese Optics, 2017, 10(3): 331-339.
[2] Song K C, Wen X, Zhao Y J. Noise robust image matching using adjacent evaluation census transform and wavelet edge joint bilateral filter in stereo vision[J]. Journal of Visual Communication and Image Representation, 2016,38(1): 487-503.
[3] Sebastián C, José-Emilio G. SIFT optimization and automation for matching images from multiple temporal sources[J]. International Journal of Applied Earth Observation and Geoinformation, 2017, 57(1):113-122.
[4] 趙雙,楊慕升. 結(jié)合小波變換的Shi-Tomasi算法遮擋圖像匹配研究[J]. 機械設(shè)計與制造, 2017, 11(1):118-121.
[5] Yang J F, Shi Y H, Jia G M. Finger-vein image matching based on adaptive curve transformation[J]. Pattern Recognition, 2017, 66(1):34-43.
[6] 傅超斌,南開來. 基于改進遺傳算法的圖像匹配定位[J]. 微型機與應(yīng)用, 2017, 36(23):44-45+49.
[7] Besbes B,Rogozan A,Rus A M. Pedestrian detection in far-infrared daytime images using a hierarchical codebook of SURF[J]. Sensors (Basel), 2015, 15(4): 8570-8594.
[8] 賀一楠,耿娟,秦軍. 結(jié)合Forstner與NCC約束的UAV圖像配準(zhǔn)研究[J]. 國土資源遙感, 2015, 27(1): 48-54.
[9] Zhang J, Luo T J, Gao G. Junction point detection algorithm for SAR image[J]. International Journal of Antennas and Propagation, 2013, 2(52):861-870.
[10] 唐亞平,陳蘇婷. 一種改進的亞像素角點提取算法[J]. 電腦知識與技術(shù), 2014, 19(10):4552-4555.
[11] 胡旻濤,彭勇,徐赟. 基于改進SURF的快速圖像配準(zhǔn)算法[J]. 傳感器與微系統(tǒng), 2017, 36(11): 151-153.
[12] Zhila B, Fardin A T. A new robust video watermarking algorithm based on SURF features and block classification [J]. Multimedia Tools and Applications, 2018, 77(1): 327-345.
[13] Li A M, Jiang W L, Yuan W H. An improved FAST plus SURF fast matching algorithm[J]. Advances in Information and Communication Technology, 2017, 107(1): 306-312.
[14] 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.
[15] Li J Y, Hu Q W, Ai M Y. Robust feature matching for geospatial images via an affine-invariant coordinate system[J]. The Photogrammetric Record, 2017, 32(159): 317-331.
[16] Zhang Y, Lu K, Gao Y H. Fast image matching algorithm based on affine invariants[J]. Journal of Central South University, 2014, 21(5): 1907-1918.
[17] Zhao C Y, Zhao H C, Lv J F. Multimodal image matching based on multimodality robust line segment descriptor[J]. Neurocomputing, 2016, 177(3):290-303.
[18] 陳虹,肖越,肖成龍. 基于SIFT算子融合最大相異系數(shù)的自適應(yīng)圖像匹配算法[J].計算機應(yīng)用, 2018, 38(5):1410-1414.
[19] 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.
[20] Dou J F, Li J X. Image matching based local delaunay triangulation and affine invariant geometric constraint[J]. Optik, 2014, 125 (1):526-531.
RESEARCH ON IMAGE MATCHING ALGORITHM BASED ON CIRCULAR REGION GRADIENT INFORMATION COUPLING ANGLE SIMILARITY RULE
ZHANG Hao
(School of Information Technology, Anhui Vocational College of City Management, Anhui, Hefei 230011, China)
In order to improve the accuracy of algorithm matching and anti-affine transformation ability, by introducing the gradient information of the image into the process of image matching, An image matching algorithm based on circular angle gradient information coupling angle similarity rule was proposed in this paper. Firstly, the feature points of the image are detected by Forstner operator. Then, the main direction was determined by calculating the Haar wavelet response value of the circle domain, and the feature vector was obtained by using the gradient information of the circle domain to complete the feature description. Finally, the angle similarity rule was established by using the main direction of the feature points to complete the feature matching. Experiments show that the proposed algorithm has high matching accuracy and robustness, which has many correct matching pairs and a small number of mismatches in the matching results under various geometric transformations.
image matching; forstner operator; circular region gradient information; Haar wavelet; angle similarity rule
1674-8085(2018)05-0050-07
TP391
A
10.3969/j.issn.1674-8085.2018.05.010
2018-05-19;
2018-08-17
安徽省高校質(zhì)量工程資助項目(2017jxtd108)
張 昊(1970-),男,安徽合肥人,副教授,碩士,主要從事圖像處理、網(wǎng)絡(luò)通信、計算機應(yīng)用等方面研究(E-mail: ZhangHao1970ah@126.com).