Numpy Rush Project

Learning how to make Instagram Filters using Python's NumPy and MatPlotLib libraries

The gist of this 2 day Rush Project was to learn about NumPy and MatPlotLib

import urllib.request as urllib2
import numpy as np
import matplotlib as matplot
import matplotlib.pyplot as plot
import matplotlib.image as mpimg
import matplotlib.colors as mpcolors

Seems like the main idea behind this project was to do some mathematical manipulations with matrices and I might have went the wrong way.

Having been exposed to working with photography and videography, I was familiar with different color spaces besides RGB such as HSL, HSV, CMYK, HEX etc. Just out of curiousity, I found out that matplotlib had a sub-library called colors, which can be used to convert the img matrix from RGB to HSV. It gives more freedom to operate with such values as Saturation and Brightness.

Reading from the Image

It's just a very basic code to read the image file. imread automatically supports PNG format. For all other formats, you have to specify.

url = 'https://eldardev.s3.us-east-2.amazonaws.com/liana.jpg'
img_readonly = mpimg.imread(urllib2.urlopen(url), format='jpeg')
img = np.copy(img_readonly)
plot.imshow(img)
plot.show()

I wrote several generic functions to be used in all of my filters:

Adjust Saturation

This functions takes the img matrix, converts it to HSV and adjusts the Saturation by perc value.

def adjust_saturation(img, perc):
  img_hsv = mpcolors.rgb_to_hsv(img)
  for i in range(0, img_hsv.shape[0]):
    for j in range(0, img_hsv.shape[1]):
        img_hsv[i,j,1] = img_hsv[i,j,1] * ((100 + perc) / 100)
  img = mpcolors.hsv_to_rgb(img_hsv)
  img = img.astype(int)
  return (img)

Adjust Brightness

This functions takes the img matrix, converts it to HSV and adjusts the Brightness by perc value. I also had to include the threshold variable. Some filters wanted to bring up the brights, so you want to separate your brights from the rest.

def adjust_light(img, perc, threshold):
  img_hsv = mpcolors.rgb_to_hsv(img)
  for i in range(0, img_hsv.shape[0]):
    for j in range(0, img_hsv.shape[1]):
      if (img_hsv[i,j,2] > (threshold / 100.)):
        img_hsv[i,j,2] = img_hsv[i,j,2] * ((100 + perc) / 100)
  img = mpcolors.hsv_to_rgb(img_hsv)
  img = img.astype(int)
  return (img)

Add Color

This functions takes the img matrix and applies the color (passed as an RGB value, eg. [255, 122, 34]) value with the perc intensity. Basically, all the pixels in the image are moving towards that color. You determine if the just move 15% of the way or all 100%)).

def add_color(img, color, perc):
  for i in range(0, img.shape[0]):
    for j in range(0, img.shape[1]):
      for k in range(0, img.shape[2]):
        diff = color[k] - img[i,j,k]
        img[i,j,k] = img[i,j,k] + diff * perc / 100
  img = img.astype(int)
  return (img)

Apply Vignette

This function applies the vignette effect. For this, I had to remember the formula of the ellipse which is

$$(x^2)/(a^2) + (y^2)/(b^2) = 1$$

So, in my case, I wanted to modify only pixels that were outside this ellipse. Therefore, I checked if this formula was >1. Also, make sure you move the x and the y to the middle of your image.

def apply_vignette(img):
  img_hsv = mpcolors.rgb_to_hsv(img)
  center_x = img.shape[0] / 2
  center_y = img.shape[1] / 2
  for i in range(0, img_hsv.shape[0]):
    for j in range(0, img_hsv.shape[1]):
      diff = ((i - center_x)**2) / (center_x**2) + ((j - center_y)**2) / (center_y**2)
      if (diff > 0.95):
        img_hsv[i,j,2] = img_hsv[i,j,2] * (1 - (diff - 0.95)**(1.2))
  img = mpcolors.hsv_to_rgb(img_hsv)
  img = img.astype(int)
  return (img)

Example

To see more, just open the Rush - Part II file. GitHub will automatically load it and run it for you, which is really nice!!

P.S. I might have done everything completely wrong. It was my first time using Python for a project.