iamkirankumaryadav/opencv

Open Computer Vision

Open Computer Vision

!pip install opencv-contrib-python
import cv2

• An open source computer vision and machine learning library.
• Contains over 2,500 optimized algorithms.
• Algorithms for Facial recognition, Object identification, Human action, Track movements,etc.
• Computers gain high level understanding from digital image and video
• Understand and automate the task of human visual system
• Detecting and labelling image or object has surpassed humans.
• Faster than human reaction and 99% Accuracy
• High amount of visual data i.e images and videos with high processing capabilities.

OpenCV and NumPy

• OpenCV code is orignally written in C++
• When we use OpenCV in Python, Python act as wrapper around C++ Code.
• All the OpenCV array structures are converted to and from NumPy arrays
• NumPy arrays are powerful N dimensional array objects.
• NumPy have sophisticated broadcasting functions.
• It makes processing for images faster.
• Every NumPy array is a grid of elements of the same type.

Image Processing

• Extracting infromation from image and enhancing the image quality.
• Computers visualize images as a matrix
• Each element in matrix is a pixel
• OpenCV reads the image in channel order BGR format but Matplotlib and other visualization tools read in RGB order.
• Each pixel have color information composed of 3 channels (B G R) containing numbers from 0 to 255
• Each pixel of matrix represents the intensity of an color.
• Each channel need 8 bits for storage therefore it become 24 bits for each pixel.
• Some images need to be converted to Grayscale or Binary before applying any image processing.
• Blending : Overlaying images on top of each other by adjusting dimensions and opacity.
• Thresholding : Segmenting or dividing images into different parts.
• Binary Thresholding : Value below threshold will be shifted to 0 else 255
• Bianry Inverse Thresholding : Value below threshold will be shifted to 255 else 0
• Truncation Thresholding : Value below threshold will be shifted to threshold value else remain same
• Threshold To Zero : Value below threshold will 0 else remain same
• Threshold To Zero Inverse : Value below threshold will remain same else 0
• Blurring : Adding noise
• Smoothing : Removing noise to focus more on general details of the images.
• Edge Detection : Detect edged in high resolution images without bluring.
• Gamma Correction : Makes images brighter or darker based on choosen gamma value ( Lower gamma value : Brighter images )
• Kernels are used for Blurring and Smoothing

Morphological Operators

• Erosion : Remove pixels on object foreground and boundaries (Shrink the boundaries)
• Dilation : Add pixels to the boundaries of object in an image (Enlarge the boundaries)
• Opening : Remove white noise from black background.
• Closing : Remove black noise from white foreground.

print(Image.shape)

(340, 540, 3) : (height, width, channel)
340 represents rows | height 
540 represents columns | width
3 represents channels.

Images are converted to and from NumPy Array

Storage Requirements

Image Type	Storage (Bits Per Pixel)	Channel
Binary	1
B/W	8	1
Grayscale	8	1
Color	24	3

Color	Channel
B	0
G	1
R	2

Pixel

• Basic fundamental unit of an image.
• Multiple pixels arranged in rows and columns forms an image.
• Binary, black and white, grayscale, RGB color image.

Black & White and Greyscale images are different

• B/W is only composed with either black or white
• Greyscale is combination of different shades of black and white.

Image Formats

• HSV
• HLS

Applications 

1. Autonomous self driving cars : Detect objects and humans infront of car, reverse parking.
2. Facial recognition : Attendance, FB friend recognition.
3. AR : Augmented Reality and VR : Virtual Reality
4. Health care : X Ray and MRI scan.
5. Video motion analysis : Detect face, Object in video.
6. Image segmentation : Camera detects the multiple faces in a group selfie.
7. Scene reconstruction : 3D model creation in architecture, room interior.
8. Image restoration : Filtering blur images and removing noise.

DNN is OpenCV's deep learning module.

The below step you will find in many articles and videos, It simply represents :

• To mirror the image you simply need to reverse either the x or y coordinates.

# Trick to reverse BGR to RGB and keep x and y as it is (One liner code)
image[:, :, ::-1] 

# You can also the method :
cv2.cvtColor(src=image, code=cv2.COLOR_BGR2RGB)