联合稀疏表示的医学图像融合及同步去噪

Simultaneous Medical Image Fusion and De-Noising with Joint Sparse Representation

将多模态医学图像的互补信息有机地融合在一起,可为临床诊断和辅助治疗提供丰富信息和有效帮助。基于联合稀疏模型,提出一种联合稀疏表示的医学图像融合算法,当图像被噪声污染时,该算法在融合的同时兼有去噪功能。首先,将配准的源图像编纂成列向量并组成联合矩阵,通过在线字典学习算法(ODL)得到该矩阵的超完备字典;其次,利用该字典得到联合稀疏模型下的联合字典,之后利用最小角回归算法(LARS)计算基于联合字典的公共稀疏系数和各图像的独特稀疏系数,并根据"选择最大化"融合规则得到融合图像的稀疏系数;最后,根据融合系数和超完备字典重构融合图像。将该算法与3种经典算法比较,结果显示其主观上亮度失真和对比度失真较小,边缘纹理清晰,客观参数指标MI、QAB/F在无噪声干扰和有噪声干扰时的统计均值分别为:3.992 3、2.896 4、2.505 5和0.658、0.552 4、0.439 6,可以为临床诊断和辅助治疗提供有效帮助。

The complementary information of multi-modality medical images can be integrated together,which can provide abundant information and effective help for clinical diagnosis and treatment. Based on the joint sparse model,a new medical image fusion algorithm based on the joint sparse representation was proposed in this paper,and this method could carry out image fusion and de-noising simultaneously while the images were corrupted by noise. First,the registered source images were compiled into column vectors and composed of a joint matrix,and then an over-complete dictionary was obtained through online dictionary learning algorithm（ ODL）. Second,a joint dictionary was obtained by the over-complete dictionary under the joint sparse model,then based on the joint dictionary,the common sparse coefficients and unique sparse coefficients were computed by the least angle regression algorithm（ LARS）,and the sparse coefficients of fused image were obtained according to the fusion rule ＂ choose max＂. Last,the fusion image was reconstructed according to the fusion coefficient and the over-complete dictionary. Compared with three classical algorithms,the results showed that the proposed algorithm has small luminance distortion,small contrast distortion and clear edge texture in the subjective vision,the statistical mean values of the objective parameters MI,QAB / Funder noiseless and noisy case were 3. 992 3,2. 896 4,2. 505 5 and 0. 658,0. 552 4,0. 439 6,respectively. All of these can provide effective help for clinical diagnosis and treatment.