Abstract：An image features extraction and fusion algorithm based on online dictionary learning (ODL) is presented in this paper. Firstly, source images were combined into a joint matrix by the sliding window technique, the size of the sliding window was 8×8, the overcomplete dictionary was trained by ODL algorithm and the sparse codes were acquired by LARS algorithm; the activity level measurement of sparse codes was the L1 norm of its vector, then, the sparse codes were fused by activity level maximum rule; finally, the fused image was reconstructed by overcomplete dictionary and fused sparse codes. Coaligned medical images of twenty patients were tested by experiments and the quality of the fused image was evaluated by five kinds of commonly used objective criterions. Compared with the other two popular medical image fusion algorithms, objective criterions of the fusion result show the advantage of the proposed algorithm, the mean of Piella, QAB/F，MIAB/F， BSSIM and space frequency index is 0.800 4, 0.552 4,
3.630 2, 0.726 9 and 31.941 3, the fusion images of the proposed algorithm have high definition and contrast, clear texture and edge and fast speed, showing its application potentials of aiding clinical diagnoses and treatment.
［1］Das S, Kundu MK. NSCTbased multimodal medical image fusion using pulsecoupled neural network and modified spatial frequency ［J］. Medical & Biological Engineering & Computing, 2012, 50(10): 1105-1114.
［2］Pajares G, Cruz J. A waveletbased image fusion tutorial ［J］. Pattern Recognition, 2004, 39(9): 1855-1872.
［3］da Cunha AL, Zhou Jianping, Do MN. The nonsubsampled contourlet transform: theory, design, and Applications ［J］. IEEE Transactions on Image Processing, 2006, 15(10): 3089-3101.
［4］Wang Zhaobin, Ma Yide, Chengk Feiyan, et al. Review of pulsecoupled neural networks ［J］. Image and Vision Computing, 2010, 28(1): 5-13.
［5］Yu Nannan, Qiu Tianshuang, Liu Wenhong. Medical image fusion based on sparse representation with KSVD ［C］// World Congress on Medical Physics and Biomedical Engineering. Beijing: SpringerVerlag, 2013, 39: 550-553.
［6］余南南, 邱天爽, 毕峰, 等. 基于KSVD的医学图像特征提取和融合［J］. 大连理工大学学报, 2012, 52(4): 605-609.
［7］Aharon M, Elad M, Bruckstein A. KSVD: An algorithm for designing overcomplete dictionaries for sparse representation ［J］. IEEE Transaction on Signal Processing, 2006, 54(11): 4311-4322.
［8］Mairal J, Bach F, Ponce J, et al. Online dictionary learning for sparse coding［C］// Proceedings of the 26th Annual International Conference on Machine Learning. New York: ACM, 2009: 689-696.
［9］Olshausen BA, Field DJ. Emergence of simplecell receptive field properties by learning a sparse code for natural images ［J］. Nature, 1996, 381(13): 607-609.
［10］Singh M, Singh S, Gupta S. An adaptive speckle suppression filter based on local statistics and edgemap for ultrasound images ［J］. Advanced Science Letters, 2013, 19(8): 2375-2379.
［11］Wang Zhou, Bovik AC. A universal image quality index ［J］. IEEE Signal Processing Letters, 2002, 9(3): 81-84.
［12］Qu Guihong, Zhang Dali, Yan Pingfang. Information measure for performance of image fusion ［J］. Electronics Letters, 2002, 38(7): 313-315.
［13］Xydeas CS, Petrovic V. Objective image fusion performance measure ［J］. Electronics Letters, 2000, 36(4): 308-309.
［14］Wang Zhou, Bovik AC, Sheikh HR, et al. Image Quality assessment: from error visibility to structural similarity ［J］. IEEE Transactions on Image Processing, 2004, 13(4): 600-612.
［15］马先喜, 彭力,徐红. 基于块同化的空间频率多聚焦图像融合算法研究［J］. 科学技术与工程, 2012, 12(1): 64-67.
