DefinePK

Abstract

Criminal Face Recognition is the ability to detect and recognize a criminal by their facial characteristics. Most crucial duty of the police searching for the offenders is criminal identification. But since the police have to look for it everywhere, it is the hardest and time-consuming duty. Cities and other public areas with a high population density will provide greater challenges. Manual identification methods occasionally provide additional information about offenders. However, there is no requirement for monitoring when using an automatic identification system (AIS) in a public setting. Because the MIS is taking longer time, it cannot adequately focus on everyone.  So, automatic identification is essential for recognizing the criminal face. A distinct machine learning method is the automated prediction model that is most frequently employed to identify criminal faces. However, attaining accurate prediction with lesser error probability is quite difficult using the existing network models. To overcome these issues, deep learning algorithm has been developed for identifying the criminals. Initially the criminal and non-criminal images are collected and pre-processed using a multi-stage progressive V-net approach. After that, the MT-CNN algorithm extracts features from the pre-processed image. The MTCNN (Multi-Task Cascaded Convolutional Networks) algorithm finds and recognizes faces in digital images or videos by employing a cascading sequence of convolutional neural networks (CNNs). Using the NASnet method, the segmented image is finally categorized to detect both criminal and non-criminal activity. The experimental analysis shows that the suggested strategy achieves 97.2% accuracy, 96.9% f1-score, and 2.60% FDR. Thus, the automated criminal detection system not only offers the police enormous convenience in identifying offenders, but it also saves them time.