LOW LIGHT IMAGE ENHANCEMENT

Image is a crucial component as a form of information with its characteristic compared to the
information only in the form of writing or text. Although an image consists of much information, the
owned image can often experience image quality reduction. For example, a too dark image will make
an unclear display of the image. The said image occurs due to insufficient or dark lighting during the
image shooting, causing an image difficult to interpret because the image is poor and it can cause its
image information to be reduced

PROBLEM STATEMENT
Develop an efficient low-light image enhancement algorithm to enhance the visibility and quality of
images taken in challenging low-light conditions without compromising important details or
introducing noise. The solution should be versatile, applicable to various image types, and maintain
colour accuracy while improving visual appeal.

GENERAL STEPS TAKEN

1. LOCALISED BRIGHTENING
Brightness is the perceived intensity of light coming from a screen. On a colour screen, it is
the average of the red, green and blue pixels on the screen. So, we extract certain areas which
are dimly lit and enhance the colour value of each pixel by a fixed amount. This makes the
colour of the region lighter and the image looks much brighter.

2. ALPHA BLENDING
Doing the above process creates a slightly blurred image. To enhance the resolution, we do
alpha blending. Alpha is the fourth variable in the colour schemes. It is the value of opacity of
the required enhanced image. This merges the image with the background providing a clearer
and brighter image.

DATA COLLECTION
This model used is a supervised learning model. Hence, various images are required and provided.
The following are some of the datasets acquired:
● Images - Various low light images of natural settings.
● Greaterset - 500 different images in low light.
● ExDark - Large set of low lit images.

MODEL

HISTOGRAM EQUALIZATION
Histogram Equalization is a computer image processing technique used to improve contrast in images.
It accomplishes this by effectively spreading out the most frequent intensity values, i.e. stretching out
the intensity range of the image. This method usually increases the global contrast of images when its
usable data is represented by close contrast values. This allows for areas of lower local contrast to
gain a higher contrast.
PARTICLE SWARM OPTIMIZER
PSO algorithm concept is the velocity vector updated on every particle, then adding up the
velocity vector to the particle position. Performing the velocity update process will be
affected by both solutions such as adjusting the best position from the particle called particle
best and adjustment on the best particle from all flocks called global best. On every iteration,
the presented solution by the particle position is evaluated by inserting the said solution into
the fitness function

ERROR CALCULATION
The accuracy calculation process on this research uses PSNR. The bigger PSNR value
indicates that the method used can increase image quality. While a smaller PSNR value
indicates a smaller enhancement in its quality. Thus, higher PSNR value is deemed better
because of the higher signal ratio against the method.

REFERENCES
● M. Tzelepi and A. Tefas, "Deep convolutional learning for Content Based Image Retrieval",
Neurocomputing, 2018.
● L. Zhuang and Y. Guan, "Image Enhancement via Subimage Histogram Equalization Based
on Mean and Variance", Comput. Intell. Neurosci, 2017.
● C. Y. Wong et al., "Histogram equalization and optimal profile compression based approach
for colour image enhancement", J. Vis. Commun. Image Represent, 2016.
● B. Chopard and M. Tomassini, "Particle swarm optimization", Natural Computing Series,
2018.
● Swagat Kumar Behera, Satyasis Mishra and Debaraj Rana, "Image Enhancement using
Accelerated Particle Swarm Optimization", Int. J. Eng. Res, 2015.