Search

In several scenarios like forensic and civilian applications, biometric has emerged as a powerful technology for person authentication. Information extracted from different biometric traits is combined by the Multimodal Biometric (MB) solutions, hence showing a high resilience against presentation attacks. Additionally, they offer enhanced biometric performance and increased population coverage that is required for executing larger-scale recognition. By employing Brownian Motion enabled K-Means Algorithm (BM-KMA) and Cosine Swish activation-based Radial Basis Function Neural Network (RBFNN) (CS-RBFNN) methodologies, an MB authentication system centered on overlapped Fingerprints (FPs), Palm Prints (PPs), and finger knuckles (FKs) is proposed here. Primarily, from the publically available datasets, the overlapped FP images and hand images are taken. Next, to separate the PPs and FKs, the Region of Interest (ROI) is estimated for the hand image. Then, pre-processing, feature extraction, and feature reduction are carried out. From the overlapped FP, the noises are removed using BF; after that, the FP's contrast is enriched using SMF-CLAHE for improving the clarity of the minutiae structure of the ridges. Following this, normalization is performed using the Min–Max operation. Minute features are extracted by separating the overlapped FP using BM-KMA, which makes the system from avoidance of system complexity by separating the overlapping. From this, interest features are selected using KRC-PCA. Next, feature fusion is conducted. Finally, CS-RBFNN is wielded to categorize genuine biometrics from imposter ones. Via performance metrics, the proposed system is further affirmed. The outcomes exhibited that the proposed technique surpasses the other prevailing methodologies. [ABSTRACT FROM AUTHOR]

Determining and proving that a specific person or several persons may or may not be there at the Crime Scene (CS) in every criminal investigation are vital. Thus, in the law enforcement community, more often the physical evidence is collected, preserved, and analyzed. The accused cannot be predicted by normal people or judge just by looking at the evidence obtained at the analysis phase. So, research studies were undertaken on automated recognition as well as retrieval system aimed at forensic Crime Scene Investigation (CSI). A survey on the investigation of forensic CS evidence is depicted here. The main focus is rendered on the computer-centered automated investigation system. The latest research on the different evidence-centered Forensic Investigation (FI), such as the face, Finger-Print (FP), shoeprint, together with other Foot-Wear (FW) impressions, Machine Learning (ML) algorithm-centered FI, ML-centered pattern recognition, features of disparate evidence in forensic CSI, and various matching technique-centered FI, is surveyed here. Finally, centered on the accuracy and other two metrics, the methods' performance for CSI is compared. Out of all the other methods, OLBP + LSSVM produced better results for precision and recall followed by CLSTM. In terms of accuracy, CLSTM produced better results than any other method. [ABSTRACT FROM AUTHOR]