Computer Vision : Part 1

As seen in the previous post, a Computer Vision is a weird combination of Biology + Math + Computer science.
Human Eye is an input layer from where the human brain receives the image data.
Processing of image using different colour models happens in the human eye (using colour models like RGB , CMYK , HSL , HSV ) human eye has a  connection with the brain and desired output is generated accordingly by Human Eye + Brain

Research topics in the area of Computer Vision in the last 50 years.




There are a series of steps involved in the implementation of any Computer Vision algorithm.
As seen below each step can be a “Business Use Case” for business men.




As seen above in the 14th step of computer vision is – “RECOGNITION"

IMAGE RECOGNITION is the latest discovery in Computer Vision which includes the following:
 Object detection
 Face recognition
 Instance recognition
 Category recognition
 Context and scene understanding
 Recognition databases and test sets

I will cover the first 2 steps in today’s post (Image formation and Image processing)
This will be my PART 1 OF COMPUTER VISION post.



Image Formation:

Types of Image data
Multi-spectral Image
Geospatial data
Clinical trial data – X Rays and Scan
Satellite imagery data
Digital data
      Etc. and many more

Like human eye in computer vision - first a vocabulary has to be established to describe the geometry of the image.

There are 2 Models for Image formation




Geometric Formation:

Geometric primitives form the basic building blocks used to describe three-dimensional shapes.
2D Points are denoted by Pixel value of the image.
Pixel values are created for the images that is stored in the database and with the help of pixel values - Points, lines, and planes are introduced into the current image database.
With the help of lines, equations are created as seen below.


Image Transformation:

After having defined our basic primitives, image now has to be transformed.
The simplest transformations occur in the 2D plane and are illustrated in below figure.
As seen below Image Transformation has following steps.



Translation: Translate the data into a matrix
Rotation: Using Euclidean distance
Transformation of the Image: Affine Transformation and Projective / Perspective Transformation
Same steps are also followed by the 3D Images for image transformation except that different rotation techniques are used by 2D and 3D Images.
Difference between 2D and 3D Image transformation can be studied more in the below link.


Below figure shows the 3D Image transformation models
The most commonly used transformation in computer graphics is Perspective Transformation






Photometric image formation:
Image cannot exists without light
A particular image is produced given a set of lighting conditions, colour, and shadow and camera optics
In the photo-metric image formation model - features like colour, shadow and light are created for the current image database


Pixel values are influenced by the following features which are introduced into the current data model
Colour  
Shadow
Light 
Surface area of the object 
Camera Optics



Image Processing:

We have seen how images are formed through the interaction of 3D scene elements,Lighting, and camera optics and sensors.
Image processing is like data cleaning stage.
In this stage the image is pre-processed and converted into database (analytical file) which can be used for further analysis.
Examples of pre-processing operations include
1. Exposure correction
2. Colour balancing
3. Reduction of image noise
4. Increase sharpness of the image
5. Straightening the image by rotating it




I will not cover all the pre-processing methods ,and discuss only the main points

Point processing:  Manipulating pixels independent of its neighbours



Fourier Transform tool: The Fourier Transform is an important image processing tool which is used to decompose an image into its sine and cosine components
This is one of the most important Image processing technique 



Global optimisation of images for further analysis using Bayesian Markov random field Models






After image processing is – “Feature detection and image segmentation” which I will cover in part 2 of computer vision post.
A glimpse of Feature detection and image segmentation
      





Comments