CN110211081A - A kind of multi-modality medical image fusion method based on image attributes and guiding filtering - Google Patents

Abstract

The multi-modality medical image fusion method based on image attributes and guiding filtering that the invention discloses a kind of, frame is decomposed first with transportable frame, and source images are decomposed into texture ingredient and proximate component, it is merged later using a kind of fusion rule pairing approximation ingredient based on image attributes and guiding filtering of design, texture ingredient is merged using maximum absolute value rule, finally rebuilds original image.The invention has the advantages that this method can retain more image borders and texture information, blending image contrast with higher simultaneously, and more meet the vision of people, since this method can not only obtain preferable syncretizing effect, computational efficiency is higher simultaneously, has potential application in multi-modality medical image fusion system.

Description

Multimode medical image fusion method based on image attribute and guide filtering

Technical Field

The invention relates to the technical field of image fusion, in particular to a multimode medical image fusion method based on image attributes and guide filtering.

Background

As is known, medical imaging plays an increasingly important role in clinical applications such as diagnosis, therapy planning, surgical navigation, etc., and due to the diversity of imaging mechanisms, multi-modal medical images in different modes are focused on different organ and tissue information. Computed Tomography (CT) can accurately detect dense structures and magnetic resonance imaging (MR) can provide high resolution anatomical information of soft tissues. Positron Emission Tomography (PET) and Single Photon Emission (SPECT) can reflect metabolic information of the body. In order to obtain sufficient information for accurate diagnosis, the multi-modality medical image fusion technique is an effective method. The method aims to fuse complementary information in a plurality of medical images in different modalities to generate a composite image.

Generally, image fusion is divided into three levels from low to high: the method comprises pixel level fusion, feature level fusion and decision level fusion, and the research of the patent is pixel level image fusion.

Current multimodal medical image fusion methods can be mainly divided into two main categories: transform domain fusion algorithms and spatial domain fusion algorithms. The transform domain fusion algorithm mainly comprises the following steps: firstly, the image is transformed to a specific image representation domain, then the image representation coefficients are fused by utilizing a fusion rule, and finally, a fused image is obtained by utilizing inverse transformation. The transform domain fusion method can generally obtain better effect in the field of image fusion. The choice of transformation and fusion rules is two fundamental problems with such approaches. In addition, this method has a problem that it is difficult to determine the number of layers to be decomposed, and it usually takes a long time due to the process of transformation and inverse transformation. The spatial domain fusion method is different from the transform domain fusion method. The spatial domain fusion method fuses source images in a spatial domain, generally, the calculation complexity of the method is low, and the patent provides a solution in the spatial domain fusion algorithm for solving the problems of the transform domain fusion algorithm.

Disclosure of Invention

In order to solve the technical problem, the invention provides a multimode medical image fusion method based on image attributes and guide filtering.

The invention adopts the following technical scheme: a multimode medical image fusion method based on image attributes and guide filtering comprises the following steps:

first, moving frame decomposition diagram (MFDF) decomposition:

decomposition of source images { A, B } into their respective texture components { A, B } Using MFDF_T,B_TAnd the respective approximate components { A }_A,B_A}；

Secondly, texture component fusion:

using the maximum absolute value rule to match the texture component { A ] of the source image_T,B_TGet the fusion texture component F by fusion_TFurthermore, to make the information of neighboring pixels from the same image as much as possible, we apply to the texture component F_TCarrying out consistency detection;

thirdly, fusing approximate components:

1. firstly, Gaussian filtering and Laplace filtering are utilized to obtain an approximate component { A_A,B_AInitial weight map of { P }_A,P_B}；

2. We then refine the initial weight map { P } using a thresholding method and a guided filtering method_A,P_BGet the final weight map { T }_A,T_B}；

3. Final fused approximate component F_AIs an approximate component { A_A,B_AAnd the final weight map { T }_A,T_B-weighted summation of;

fourthly, image reconstruction:

blending texture components F_TAnd the fusion approximation component F_AThe arithmetic square root of the sum of squares of the pixels at the corresponding positions is the pixel value at the corresponding position of the fused image F.

Compared with the prior art, the invention has the advantages that: the method can keep more image edge and texture information, the fused image has higher contrast ratio and better accords with human vision, and the method has potential application value in a multimode medical image fusion system because the method can obtain better fusion effect and has higher calculation efficiency.

Drawings

Fig. 1 is a basic block diagram of the multimodal medical image fusion method of the present invention.

Detailed Description

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, and the scope of the present invention will be more clearly and clearly defined.

Examples

Step 1 Moving Frame Decomposition Framework (MFDF) decomposition

The moving frame decomposition framework can decompose an image I into texture components I_TAnd approximate component I_A. The method comprises the following basic steps:

1. creating a matrix P that can be used for image decomposition

Representing the gradient of the image I, I_x and I_yRepresenting the horizontal and vertical partial derivatives of the image I, respectively. Mu is a set constant, the value of which is chosen to have some influence on the decomposition result.

2. Obtaining texture component I of image I according to formula (2) by using matrix P_TAnd approximate component I_A

In this patent, we decompose the source images { A, B } into their respective texture components { A, B } using equations (1-2)_T,B_TAnd the respective approximate components { A }_D,B_D}。

Step 2 texture component fusion

Using the maximum absolute value rule to match the texture component { A ] of the source image_T,B_TGet the fusion texture component F by fusion_T. Furthermore, to make the information of neighboring pixels from the same image as much as possible, we apply to the texture component F_TAnd (6) carrying out consistency detection. The fusion process is represented by formula (3-5).

MAX_X＝MAJORITY(abs(X_T),W_r) (3)

mm＝((MAX_A＞MAX_B)*W)＞floor(r×r/2) (4)

F_T＝mm.×A_T+((～mm).×B_T) (5)

In equation (3-5), X ∈ { A, B }, MAJORITY denotes the MAJORITY filter function, abs is the absolute value function, W_rIs a filter template of r x r. In all formulas of this patent,. denotes convolution operation,. mm is a texture component weight map,. times.denotes dot product operation.

Step 3 approximate component fusion

In the guided filtering theory, in a local window w centered on a pixel k_kThere is a linear relationship between the filtered output O and the guide image I.

wherein Parameter a_k and b_kThe definition is as follows:

wherein ,μ_k，δ_k,|w| and and a local window w_kIt is related. In particular, mu_k and δ_kRespectively represent the guide image I in the local window w_kMean and variance of the interior, | w | is the local window w_kThe total number of pixels within the array of pixels,in a local window w for an input image P_kAverage value of (d). For convenience, we mathematically represent the guided filtering as equation (9).

O＝GF_r,ε(P,I) (9)

wherein GF_r,εRepresenting the pilot filter function, and the two sub-indices r and epsilon represent the size and blur level of the pilot filter, respectively.

In this patent, the approximate component fusion is performed in three steps:

1. first, an approximate component { A } is obtained from the equation (10-12) by Gaussian filtering and Laplace filtering_A,B_AInitial weight map of { P }_A,P_B}。

P_A＝S_A＞S_B (11)

P_B＝S_A＜S_B (12)

In the formula (10-12), X ∈ { A, B }, Lap denotes a Laplace filter function, W_lA laplacian filter template of l x l, Gau denotes a gaussian filter function,is a gaussian filtering template.

2. We then refine the initial weight map { P } using a thresholding method and a guided filtering method_A,P_BGet the final weight value map { T }_A,T_BWhich can be represented by the formula (13-15) }

w_X＝Thsegment(X_A,th_X) (13)

T_A＝GF_r,ε(((P_A|w_A)&(～w_B)),A_A) (14)

T_B＝GF_r,ε(((P_B|w_B)&(～w_A)),B_A) (15)

In the equation (13-15), X ∈ { A, B }, Thsegment is the threshold segmentation function, threshold th_XAs an image X_AMean, w, of the first 5% (pixel value size) pixels_XAs an image X_AIs divided into binary images, GF_r,εTo guide the filter function, the two sub-indices r and epsilon are set to 4 and 0.3, respectively.

3. Final fused approximate component F_AObtained according to formula (16)

F_A＝T_A.×A_A+T_B.×B_A (16)

Step 4: image reconstruction

The fused image F is obtained according to the formula (17)

Without being limited thereto, any changes or substitutions that are not thought of through the inventive work should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope defined by the claims.

Claims (1)

1. A multimode medical image fusion method based on image attributes and guide filtering is characterized by comprising the following steps:

first, moving frame decomposition diagram (MFDF) decomposition:

decomposition of source images { A, B } into their respective texture components { A, B } Using MFDF_T,B_TAnd the respective approximate components { A }_A,B_A}；

Secondly, texture component fusion:

using absolute maximum rule to source graphTexture component of image { A_T,B_TGet the fusion texture component F by fusion_TFurthermore, to make the information of neighboring pixels from the same image as much as possible, we apply to the texture component F_TCarrying out consistency detection;

thirdly, fusing approximate components:

1. firstly, Gaussian filtering and Laplace filtering are utilized to obtain an approximate component { A_A,B_AInitial weight map of { P }_A,P_B}；

2. We then refine the initial weight map { P } using a thresholding method and a guided filtering method_A,P_BGet the final weight map { T }_A,T_B}；

3. Final fused approximate component F_AIs an approximate component { A_A,B_AAnd the final weight map { T }_A,T_B-weighted summation of;

fourthly, image reconstruction:

blending texture components F_TAnd the fusion approximation component F_AThe arithmetic square root of the sum of squares of the pixels at the corresponding positions is the pixel value at the corresponding position of the fused image F.