Registration of two space closed Images

Algorithms
1

Dear All,

Happens that I take the Region of Interest of two images. Those images are space closed. I apply a Registration in Rotation and Translation (Mutual Information) and with the registration the result is worst than without registration.

How I can deal with this? Low learning rate?

I want the same systematic registration for apart ROIs and for a close ones.

Thanks,

Luis Gonçalves

2

Worse how? What are you trying to accomplish?

3

The two images, instead of become more close together, they come more apart.

4

Maybe your applying the resulting transform in the wrong direction? When applying to moving image, use direct transform. When applying to fixed image, use inverse transform.

If this is not the issue, we will need more detailed description and preferably some screenshots showing what you do.

5

Sir,

Someone inform me the contrary in the following post. I suppose. Please see.

https://discourse.itk.org/t/problems-with-registration-fusion/3118/5

Thanks,

Luis Gonçalves

6

So that is not the issue.

We will need more detailed description and preferably some screenshots showing what you do.

7

I have an working example that is working perfectly.

But the others no.

Please, see better the other post. He is saying the contrary from you.

Transform

Dimensions=0.3 in all directions

[0.0068103056, -0.049634278, 0.050176658, 1.9247658, 3.810626, 2.4928436]

PreRegistration Images (ROI)

https://www.luisgo.pro/down/pre_regis1.png

https://www.luisgo.pro/down/pre_regis2.png

PosRegistration Images (ROI)

https://www.luisgo.pro/down/pos_regis1.png

The point in the corner is (0,0,0). If the faces are facing you, to inside the screen is Z, for downwards is X and for the side is Y.

The image with flat nose is the fixed image.

https://www.luisgo.pro/down/Total_Image_Registration.png

In the result registration image the inside has several levels of grey or color that was applied a threshold to attest the good or not good registration.

8

I am now exchanging the fixed image with the moving image and changing the signal of the transform. I supposing that you are right and the other post not.

9

Problem solved. Rotation X->Y->Z and not Z->Y->X.