Masks

1. Masks

You can restrict your analysis to only the most important parts of an image by creating and applying image masks.

2. Masks

A mask is a Boolean array which serves as a screen to remove undesirable pixels. Masks must retain the same dimensions as the original image so that the two can be overlaid.

3. Creating masks

One way to create masks is to find all pixels in an image that satisfy a certain condition. For example, let's create a three by three array of ascending values. If we test for values greater than 5, we will return a three by three array where the values are True when they greater than 5, and False when they are less. Logical operations include comparisons and tests of equivalence. You can also chain operations together to select a specific range of pixels.

4. Creating masks

Let's look at this in action. Recall that the foot x-ray we have been working with has an intensity distribution like this. We see that there is a steep drop-off around 32, so let's select values greater than this. This seems to do quite a good job highlighting the foot.

5. Creating masks

Bone is the highest intensity tissue in an x-ray, and if we increase our threshold to 64, we create a rough bone mask. Finally, we can create a mask of non-bone tissue by finding pixels that are in mask 1 and not in mask 2. The selected pixels are in the foot but are not part of the bone. They seem to be related to skin and other tissue.

6. Applying masks

Masks can be used to screen images, allowing the original values through except where the mask evaluates to False. NumPy's where() function is useful for this purpose. where() applies a condition on each pixel, and instead of returning a Boolean, it returns x when True and y when False. Each of the arguments can be either arrays or single values, allowing for great flexibility. To see this in action, first import NumPy. Let's try to filter out pixels that are not part of the bone. We'll call "where im is greater than 64, return im, otherwise return 0”. Plotting the masked image shows that only the high-intensity values remain, and these are mostly bone.

7. Tuning masks

Data is noisy, so your masks will rarely be perfect. Fortunately, there are simple ways to improve them. To increase the size of your mask, you can add pixels around the edges, a process known as dilation. This can help when the edges are fuzzy or to make sure you don't accidentally mask out pixels you actually care about. To do this, we call the binary_dilation() function, which converts all background pixels adjacent to the mask into mask pixels.

8. Tuning masks

The opposite operation, "binary erosion" can be implemented in the same manner. Use it to cut the mask down to its more central pixels. You can perform these tuning operations many iterations to make your mask much larger or smaller. You can also combine the operations to open or close holes in your mask.

9. Let's practice!

You've learned the basics of masking. Now let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.