Intelligent Criminal Face Recognition Using Deep Feature Refinement and Adaptive Machine Learning for Enhanced Identification Accuracy

Abstract

The identification of individuals involved in criminal activities through facial recognition technology presents considerable difficulties stemming from inconsistencies in illumination, head orientation, obstructions, and changes in facial demeanour. This study introduces an advanced system designed to overcome these obstacles, utilizing convolution neural networks (CNNs) for the extraction of complex facial attributes and employing adaptable learning methods for precise subject verification. Initially, incoming images undergo automated preparation, involving techniques such as histogram manipulation, landmark-guided orientation, and signal enhancement, to promote feature uniformity. The CNN-based feature extraction mechanism captures both overarching and localized facial characteristics, while a feature enhancement component, incorporating principal component analysis (PCA) and emphasis-driven prioritization, amplifies discriminatory capability. An adaptable verification algorithm modifies similarity criteria based on variability within subject categories to ensure dependable identification across diverse data collections. Assessment using publicly accessible repositories of criminal facial images indicates that the presented methodology achieves enhanced identification precision, exhibiting advancements of up to 12% compared to conventional CNN-based recognition systems, dependable function in the presence of obstructions, and diminished incorrect positive identifications. These findings substantiate that the integration of sophisticated feature enhancement with adaptable machine learning provides a trustworthy and effective resolution for practical criminal identification and monitoring endeavours.