How to make a high ISO effect in a dark room? - algorithm

Question

When we use mobile or semiprofessional cameras in the room with bad light they increase ISO as a rule, the result is shown below:
This is a frame from video, as you can see there is a lot of noise. Maybe it's a little weird but I need to generate a similar noise on high quality video. However, a simple noise generator will produce a something like that:
Does anyone have some ideas how to get result like on the first frame? Or maybe there is a some existing noise generator/algorithm to make it? I will be grateful for any help.

Answer 1

I assume you simply added small random RGB to your image...
I would leave color as is and try to reduce color intensity by small random amount to darken the image and create a similar form of noise. Just by multiply your RGB color with random number less than one ...
To improve result visual the random value should be with gauss distribution or similar. Here small C++/VCL example:
//$$---- Form CPP ----
//---------------------------------------------------------------------------
#include <vcl.h>
#include <math.h>
#pragma hdrstop
#include "win_main.h"
//---------------------------------------------------------------------------
#pragma package(smart_init)
#pragma resource "*.dfm"
TMain *Main;
Graphics::TBitmap *bmp0,*bmp1;
//---------------------------------------------------------------------------
__fastcall TMain::TMain(TComponent* Owner) : TForm(Owner)
{
// init bmps and load from file
bmp0=new Graphics::TBitmap;
bmp1=new Graphics::TBitmap;
bmp0->LoadFromFile("in.bmp");
bmp0->HandleType=bmDIB;
bmp0->PixelFormat=pf32bit;
bmp1->Assign(bmp0);
ClientWidth=bmp0->Width;
ClientHeight=bmp0->Height;
Randomize();
}
//---------------------------------------------------------------------------
void __fastcall TMain::FormDestroy(TObject *Sender)
{
// free bmps before exit
delete bmp0;
delete bmp1;
}
//---------------------------------------------------------------------------
void __fastcall TMain::tim_updateTimer(TObject *Sender)
{
// skip if App not yet initialized
if (bmp0==NULL) return;
if (bmp1==NULL) return;
int x,y,i,a;
union _color
{
BYTE db[4];
DWORD dd;
};
// copy bmp0 into bmp1 with light reduction and noise
for (y=0;y<bmp0->Height;y++)
{
_color *p0=(_color*)bmp0->ScanLine[y];
_color *p1=(_color*)bmp1->ScanLine[y];
for (x=0;x<bmp0->Width;x++)
{
p1[x]=p0[x];
for (a=40,i=0;i<10;i++) a+=Random(10); // "gauss" PRNG in range 40 .. 140
for (i=0;i<3;i++) p1[x].db[i]=(DWORD(p1[x].db[i])*a)>>8; // multiply RGB by a/256
}
}
// render frame on App canvas
Canvas->Draw(0,0,bmp1);
// bmp1->SaveToFile("out.bmp");
}
//---------------------------------------------------------------------------
input image:
output image:
You can play with the PRNG properties to tweak light and noissyness.

Answer 2

What you see in that frame is not the raw sensor noise, nor a simulation of film grain as the accepted answer seems to suggest. Instead, it is the result of a noise-reduction filter that is applied to the high-ISO image. Without the filter, you’d see a lot of noise, mostly Poisson noise.
I don’t know what noise-reduction filter is built into the camera, but it likely is applied to the raw image, before conversion to RGB. Here are many papers describing such filters.

Answer 3

You basically need to increase "grain" size, and maybe flat out your noise spots. It will be hard to obtain a natural-looking result, as those grain spots in a video are obtained from various types of interpolations over values obtained from the camera sensor (with specific noise that a sensor produces). Take for example "analog" film cameras (the pictures taken with the film have that grainy natural look from the effective grains sizes/shapes of minerals that are used in the film itself). If it were so easy to produce a natural-looking film alike filter over digital images the film industry would not see the comeback that happens now.
With that said there are three things that come to my mind that could work:
Median Blur over (image + some noise): https://docs.opencv.org/master/d4/d13/tutorial_py_filtering.html
An image that has is generated with Perlin noise, scaled accordingly to your specific frame size + added over colors of your image * some factor:
https://github.com/ruslangrimov/perlin-noise-python-numpy
* factor + your image = result
Film a really dark room with a camera set up to produce the same frame resolution and with high iso, add obtained video frames over your video (some image manipulation to be applied).
Hope any of this helps and good luck with your project.