Abstract:
Criminal identification from a blurry image is a challenging task from a video scene.
Many criminals are unidentified due to blur degradation or additional noise in videos.
This paper introduces a novel approach of identifying the perpetrators using conditional
Generative Adversarial Network. This network figures out how to distinguish the
criminals, yet additionally, it learns a loss function for training the model. In this method,
our model evaluates from a self-created dataset where the video frames are blurred or
unclear. This work demonstrates how to detect criminals from the blurred dataset using
cGAN. The conditional model of GAN makes the generator stronger to produce clear
images from unclear data after a specific number of iterations. Then, the trained stable
discriminator able to distinguish the real criminal. The proposed method performs
significantly in an efficient way to identify the criminal which might be a revolutionary
approach for the welfare of the society.
Description:
In our society, crime has been increasing widely day by day in different ways such as
snatching fighting, stealing, and so on. In many countries, the problem is enhancing
alarmingly. These sorts of criminal activities have been happening frequently in populous
cities of underdeveloped countries. Not only undeveloped countries but also the luxury
cities of many developed countries have vouched for these kinds of violence. Street
fighting, snatching from moving vehicles like motorbike or bicycle often occurs in the
midtown of glamorous cities. There are a lot of paradigms of the street fighting where
people die after involving outrages.
In forensic science for criminal identification, images and videos play a significant
preamble in identifying criminals. In most cases, video surveillance like CCTV cameras
have been mainly used in the domain of forensic identification. Moreover, it has also
been easier to capture digital photos by the on spot eyewitnesses. The law enforcement
personnel uses footage for tracking, identify and arrest criminals.
However, it is difficult to identify the criminal at times; for producing blurred images due
to motion capture, low resolution, slow shutter speed or something else in digital devices.
Furthermore, when a crime occurs at night, the night vision CCTV footage produces
unclear or hazy images that are barely identifiable. In that case, it is not unrecognizable
that a criminal is committing a crime, but the face of the criminal is not identifiable due
to vague images. Furthermore, detecting perpetrators get tough even in daylight by video
surveillance. For instance, Britain Metropolitan police revealed that fewer than one crime
is disposed of among 1000 CCTV cameras in the capital due to insufficient quality of
images.