Nate's Blog

All images used in this post are from the amazing Unsplash.com

Introduction

We'll be making a histogram using matplotlib to display light distribution of pixel count in JPG images. Each pixel has an RGB value(red, green, blue) ranging 0 to 255, with the light value representing the sum of those values. (0,0,0) is black - zero light, and (255,255,255) is white - full light. Our x axis range will be 0 to 765.

For example - The light distribution of the this image ...

<br>

is this -

We can see a large distribution of dark pixels than light ones.

Why are we doing this? Because we can! While I don't have a ton of specific use cases for this, being able to use data to answer questions is important. Our initial question is "What is the light distribution of this image?"

What we'll be doing?

All of the following steps are in Python.

1. Use PIL to load an image into memory.
2. Shrink the image down to a pixel size we can more easily view.
3. Use numpy to convert our image into an array. Flatten the 3d array into a 2d array of the RGB values.
4. Convert the pixel array into an array of the pixel light values - the sun of the rgb values.
5. Use matplotlib to generate the histogram.

Let's get started!

---

Use PIL to load an image into memory.

PIL is an absolutely magical package for image processing. I created the getImageFromUrl(url) method that takes in a url, uses python's requests package to make the https request, and then load the image. We need to pass the response content into BytesIO to read the requests content into a format that PIL can consume and convert into an Image object.

By the end of this code, we have an image from the internet in memory as a PIL.Image object.

1from PIL import Image
2import requests
3
4def getImageFromUrl(url):
5    response = requests.get(url)
6    return Image.open(BytesIO(response.content))
7
8imageUrl = "https://images.unsplash.com/photo-1583364481915-dacea3e06d18?ixlib=rb-1.2.1&ixid=eyJhcHBfaWQiOjEyMDd9&auto=format&fit=crop&w=600&q=80"
9
10image = getImageFromUrl(imageUrl)
11
python

---

Shrink the image down to a pixel size we can more easily view.

I created a helper method to resize the image file so that it's largest side is a pixel count we pass in. This is to keep pixel count low enough to analyze quickly and in a controlled way. By the end of this block, we have a resized image with 150 pixels as the largest side, and the aspect ratio remaining the same.

1def resize_setLargestSide(image,maxSide):
2    width,height = image.size
3    widthRatio = width / (width + height)
4    heightRatio = height / (width + height)
5    if width > height:
6        newWidth = maxSide
7        widthPlusHeight = newWidth / widthRatio
8        newHeight = widthPlusHeight - newWidth
9    else:
10        newHeight = maxSide
11        widthPlusHeight = newHeight / heightRatio
12        newWidth = widthPlusHeight - newHeight
13    return image.resize((int(newWidth),int(newHeight)))
14
15newImage = resize_setLargestSide(image,150)
16
python

---

Use numpy to convert our image into an array. Flatten the 3d array into a 2d array of the RGB values.

the np.array method converts a PIL.Image object to a 3d np array - height by width by pixels (r,g,b). numpy arrays have the property shape, which in the case below returns the width, height, and 3, which is the length of the pixel. I create flattenedShape which will be used to convert the 3d array into a 2d array by multiplying the length by width, which is then passed into reshape(), a method that lives on the np array.

reshape() only works if the number of values remains the same, so had I not multiplied width by height, reshape() would have failed.

1import numpy as np
2
3imageArray = np.array(newImage)
4shape = imageArray.shape
5flattenedShape = (shape[0] * shape[1],shape[2])
6reshapedImage = imageArray.reshape(flattenedShape)
7
python

---

Convert the pixel array into an array of the pixel light values - the sun of the rgb values.

Boy do I love list comprehensions. Below takes the 2d array and converts it to a 1 dimensional array of pixel light values, by summing the 3 values of the pixel. At this point, we have our data ready to graph!

1lightValues = [sum(pixel) for pixel in reshapedImage]
2
python

---

Use matplotlib to generate the histogram.

And now, we graph!

1import matplotlib.pyplot as plt
2
3plt.hist(lightValues, bins=20, facecolor = 'blue')
4plt.ylabel("Amount of Light")
5plt.xlabel("Pixel Concentration")
6plt.title('Light Values')
7plt.axis([0,775,0,4000])
8plt.show()
9
python

Full Code

1from PIL import Image
2from io import BytesIO
3import requests
4
5def getImageFromUrl(url):
6    response = requests.get(url)
7    return Image.open(BytesIO(response.content))
8
9imageUrl = "https://images.unsplash.com/photo-1583364481915-dacea3e06d18?ixlib=rb-1.2.1&ixid=eyJhcHBfaWQiOjEyMDd9&auto=format&fit=crop&w=600&q=80"
10
11image = getImageFromUrl(imageUrl)
12
13def resize_setLargestSide(image,maxSide):
14    width,height = image.size
15    widthRatio = width / (width + height)
16    heightRatio = height / (width + height)
17    if width > height:
18        newWidth = maxSide
19        widthPlusHeight = newWidth / widthRatio
20        newHeight = widthPlusHeight - newWidth
21    else:
22        newHeight = maxSide
23        widthPlusHeight = newHeight / heightRatio
24        newWidth = widthPlusHeight - newHeight
25    return image.resize((int(newWidth),int(newHeight)))
26
27newImage = resize_setLargestSide(image,150)
28
29import numpy as np
30
31imageArray = np.array(newImage)
32shape = imageArray.shape
33flattenedShape = (shape[0] * shape[1],shape[2])
34reshapedImage = imageArray.reshape(flattenedShape)
35
36lightValues = [sum(pixel) for pixel in reshapedImage]
37
38import matplotlib.pyplot as plt
39
40plt.hist(lightValues, bins=20, facecolor = 'blue')
41plt.ylabel("Amount of Light")
42plt.xlabel("Pixel Concentration")
43plt.title('Light Values')
44plt.axis([0,775,0,4000])
45plt.show()
46
python

Example Outputs

input

output

---

input

output

---

input

output