Medical Image Fusion Based On Improved Bidimensional Empirical Mode Decomposition

Medical image fusion uses image information technology to get fusion imagewith more complete information from multiple source medical images afterinformation processing. Medical images of different image principles performdifferent functions. It is very difficult for a single image to give accurate medicaldiagnosis. The fused images can synthesize the complementary information ofmedical images of different image principles and thus provide more powerful basisfor the doctor’s medical diagnosis.In1996, Norden.E.Huang et proposed the Empirical Mode Decomposition(EMD). EMD is a non-parameters driver analysis tool. It can analyze nonlinearity andnon stationary signal and overcome the shortcomings that the wavelet decompositionneed to set wavelet function in advance and that the selection of wavelet function hasgreat influence on the data. EMD is a tool that is more suitable for the analysis ofnonlinear and non stationary signals. The idea of EMD is to decompose signal into aseries of Intrinsic Mode Function (IMF) and a residual. EMD decomposition processis a shift process. The IMF obtained accords with the nature of being from highfrequency to the low frequency, which can reflect the characteristics of signals well.EMD is applied to a series of problems that need high resolution but time-frequency isseparable, such as ocean engineering, bearing breakdown diagnosis, earthquake andbiomedical signal processing etc.The idea of EMD has been applied to bidimensional image analysis successfully,that is Bidimensional Empirical Mode Decomposition (BEMD). BEMD is an adaptivedecomposing process, decomposing a series of Bidimensional Intrinsic Mode Function (BIMF) from high frequency to the low frequency and a residual. There arethree key problems in BEMD: the first one is the selection of extreme value points.The selection of maximum value and minimum value plays a decisive role in thestructure of enveloping surface; the second one is the selection of interpolationfunction of enveloping surface. The common methods include triangulations andinterpolation method based on radial basis function (RBF); the third one is the stopstandard of screening. BEMD is applied in image features extraction, imagede-noising, image segmentation and image fusion etc.BEMD applied in image fusion has obtained very good effect, but there are alsosome problems. BEMD is a data driven adaptive decomposing process. The numberof decomposed BIMF of each image is different because of its own data. But theimage fusion needs the same BIMF; it is found in the process of decomposition that asingle BIMF sometimes isn’t a good response of texture characteristics of the originalimage. More reasonable method is needed to process BIMF to make BIMF can betterreflect the characteristics of the image; after different images go through BEMD, thecorresponding BIMF is not necessarily the corresponding frequency component. Inview of the above disadvantages in BEMD image fusion, this paper proposed theconcept of m-BIMF, namely, a new BIMF constructed by many BIMF together.m-BIMF has greater scale than the original BIMF, which not only reduces the numberof BIMF but also has better texture features than the original one. The significance ofraising such m-BIMF fusion lies in that reasonable m-BIMF definition can make thecorresponding component can better correspond.The paper focuses on the application of BEMD in medical image fusion, and amethod of medical image fusion based on m-BIMF is proposed. Pulse CoupledNeural Network (PCNN) has achieved very good effect in image fusion. Theimproved m-PCNN is improved in image fusion computing speed and fusion effects.Medical image fusion method based on BEMD and m-PCNN is proposed in this paper.The experimental results show that the new medical image fusion method proposed in the paper has better effect than traditional wavelet fusion, fusion pyramid method andsimple m-PCNN.