Let's talk science for a bit shall we? 
![image.png](https://gateway.ipfs.io/ipfs/QmSQ8GHKbTioALanPGL6SggVUwtv3tm4dEJfs75kDLRHym)
[source](https://www.tomtop.com/p-d5172b.html)
Imagine you're a lawyer defending a case in court and a digital image was filled as evidenced against your client. How would you verify the authenticity of that image? 

That, my friend, is where **Image Forensics** comes in. 
This problem of image authentication has been studied quite alot over the years and a lot of methods have been developed. One of such methods is to determine which camera was responsible for that image, or better yet to prove that a particular camera is responsible for the image. And this is what is referred to as image source identification. 

Fascinating right? I thought so too. 

I recently did a little review of some methodologies for Image Source Identification and i gotta say i was pretty intrigued. So much so I've  decided to base my Masters Thesis on the subject. 

# Methodologies
Firstly, let me briefly describe the image formation process in digital cameras. 
![image.png](https://gateway.ipfs.io/ipfs/QmYrHBVCSg6ZotJmUzrcaN4QNG3eKY5S4TfnqjsP2ueCrA)
Source: Bayram, S., Sencar, H. T., & Memon, N. (2008). Classification of digital camera-models based on demosaicing artifacts. Digital Investigation, 5(1–2), 49–59. https://doi.org/10.1016/j.diin.2008.06.004

When a photo is taken, light enters the lens and is passed through a series of filters, the most important been an **anti-aliasing filter**. This light is then focused onto an array of **Charge-coupled devices (CCD)** or pixels in lay man's terms. The CCD obtains the electric signal representation of the photo been taken. all this takes place in the **sensor** masked with another filter called the **Color Filter Array.**
Each pixel in an image contains three components, Red,, Green, Blue  which require separate CCD but due to the cost only one CCD is applied per pixel which captures only one component. The values for the other two channels are then **Interpolated** (guessed) using a **Demosaicing algorithm** based off of the pixels around it.
After demosaicing, other processes such as Gamma Correction and Noise Reduction are caried out and the image is saved to memory.

Now, in my study  i discovered two major methodologies.

- Due to imperfections of some of the materials used in manufacturing sensors, certain noise patterns are introduced into an image. One of which is the **Photo Response Non-Uniformity (PRNU)** which is unique across even different models of the same brand. It is sometimes referred to as a finger print. This can be used to determine the source of an image.

- During Interpolation, Demosaicing algorithms introduce correlations between the different color channels in order to correctly interpolate the color components. This leaves certain traces on the images which can be used as a way to determine the source of an image. 

Now This is just a very brief summary of these methodologies and i am looking to do some more research in the area of using PRNU based methodologies. Most methods i have seen use Unsupervised learning algorithms like Correlation Clustering and Support Vector Machines.. So i want to try something simple like Logistic regression.


I sincerely hope this wasn't a boring read for you. At least you've learnt something new.
Thank you for Reading.