ImageProcessing

You are always welcome to optimize or improve any resource in this repository by following these instructions.

---

Images: 2-D representation of the visible light spectrum

Human visual system(eye & visual cortex) is incredibly good at image processing

Color Spaces: different method to store image in computer (eg, RGB, HSV, CMYK)

OpenCV: Open source Computer Vision Software for image processing, OpenCV default color space is RGB, but stores color in BGR format, to improve processing of image operation

HSV: hue, saturation and value/brightness , is a color space that attempts to represent colors the way humans perceive it, in CV we use for color segmentation

Installation

Install OpenCV by Anaconda: conda install opencv

Install by python installer pip: pip install opencv-python

Adding filters

We can add inbuilt filters or create custom filters by ourself, We'll incorporate both here.

Original Image:

Now let's add filters to it.

Import module opencv

>>> import cv2

OpenCV load image from storage and convert into numpy array:

>>> photo = cv2.imread('imagename.jpg')

We don't need to load numpy module, opencv call its internal property.

Convert image into gray image:

>>> gray_image = cv2.cvtColor(photo , cv2.COLOR_BGR2GRAY)

Show image, but we have use waitKey() and destroyAllWindows() with imshow(), otherwise program will hang up

>>> cv2.imshow('image title', photo)
>>> cv2.waitKey()
>>> cv2.destroyAllWindows()

waitkey(): used to hold windows for some time interval, and also wait for any key input from keyword, so listen key and then go to next statement

destroyAllWindows(): close any window, open by imshow()

A very faster alternative of adding filter by inbuilt functions is by doing it manually in an optimized way. We can convert the image to gray while reading only by a simple trick given below.

>>> grey_photo = cv2.imread('image.jpg' , 0)

Remove or Amplify a color from image

Split Color from colorful image:

>>> B, G, R = cv2.split(photo)

>>> import numpy as np
>>> height = photo.shape[0]
>>> width = photo.shape[1]
>>> zero = np.zeros((height, width))

>>> cv2.merge([zero, G, R])

It give error because, array of image require datatype “unit8” but zero function create datatype float

>>>B.dtype
dtype('uint8')

What is uint, 8-bit unsigned, here 8 is 8 bit, means 2^8 = 0-255 number we can store in it.

So if u store “1000”, it won't store this: >>> photo[0][0][0] = "1000"

>>> zero.dtype
dtype('float64')

>>> zero_unit8 = zero.astype('uint8')
>>> new_photo_B_removed = cv2.merge([zero_unit8, G, R])

>>> cv2.imshow('image title', new_photo_B_removed)
>>> cv2.waitKey()
>>> cv2.destroyAllWindows()

If we want to amplify any color in image, let say green color :

>>> new_photo_B_removed = cv2.merge([zero_unit8, G+100, R])

Same thing done by the code explained above can be done in one line like this:

If we want to remove all the colour except Red

>>> photo[:,:,0:2] = 0

Custom filters

I created a custom filter similar to "ambience" in popular photo editing applications.

Just amplified BLUE and GREEN by 100 without changing RED.