X-ray Image Enhancement

The original images taken by X-ray are generally vague, and these blurred details are very useful in medical applications. Therefore, the enhancement of X-ray images has always been the focus of people's research. Next, we will introduce a very useful enhancement method: Gauss Laplacian Pyramid algorithm.

Many documents have introduced and elaborated this method, but it is generally obscure. I have made a clear implementation step for your reference. How to enhance the detailed decomposition images (as mentioned later) is not the focus of this Article. Here, we will only provide some basic introductions and have more demanding friends, you can check the professional literature to achieve this.

First, check the effect:

Original Image

Enhancement

Steps:

I. Image Decomposition:

1. Input the original image I. The size is M * n. Because the algorithm is pyramid decomposition, we set the image size to M * n as the first layer;
2. Perform 5*5 Gaussian filter on I. The result is F;
3. F is used. The specific operation is to collect F rows and columns to obtain the image R. The size is m/2 * n/2;
4. Perform 5*5 Gaussian filtering on image R, and compress the image interpolation to M * n to obtain image E;
5. Image I minus image E, and the detailed image D is obtained. The size is M * n. Here we need to enhance the detailed image to make the subsequent reconstruction image an enhanced image;
6. Assign the R value of the next sample image in step 3 to I in step 1, update m to M/2, and update n to n/2, and then execute 1 ~ 6 until the f image size is 1*1.

Part 2 reconstruction

1. Set the R size of the next sample image on the last layer (K layer) to M * n, perform Gaussian filtering on the R, and sample the image E, the size is changed to 2 * m * 2 * n, Set 2 * m = m, 2 * n = N;
2. Add e and the detail image D of the decomposed K-1 layer (because it is the K-1 layer, so its size is also 2 * m * 2 * n, 2 * m * 2 * n enhancement image en is obtained;
3. Assign the en value to the r of the first step, and then execute 1 ~ 2. Update m to M and N to n. Until the detailed image is added to the first layer to obtain the final reconstruction image.

Note that, if we do not enhance the detailed image in the decomposition process according to the above decomposition and reconstruction steps, the reconstructed image and the original input image should be exactly the same.

3. Detailed Image Enhancement Method

The detailed image enhancement method consists of two parts: one is to give different gain weights to the detailed image features of each layer in the decomposition process. In addition, for detailed image features, to ensure that the final reconstruction image is not distorted, I have adopted the exponential function to enhance it:

　　

Parameter a sets different gain weights based on different layers and displays different enhancement effects of different parameters p.

 

In the process of image decomposition, it is found that the fourth layer is a watershed. The first three layers of images have little details, and the subsequent details are relatively obvious, therefore, you can multiply the first three layers of images by a large coefficient, and multiply the following layers by a relatively small coefficient.

The detailed image enhancement is described in more detail in the relevant literature. In addition, you can adjust the illumination of the image after the Gauss convolution to make the final reconstruction effect more perfect.