Paper Title
A Hybrid Deep Learning Approach for Forensic Facial Sketch Generation and Matching

Abstract
Witness-based suspect portrait generation is a cornerstone of criminal investigation when no surveillance imagery or photographic record can be retrieved. Reliance on specialist artists, however, renders this process both time-consuming and inconsistent. To overcome these constraints, we engineer a web-native identification platform that fuses a guided portrait composition interface with a convolutional neural network retrieval engine. Portraits are assembled from catalogued photographic facial-component crops through a React drag-and-drop workspace, or submitted as pre-existing hand-drawn files. FaceNet512 invoked through the Deep Face library converts each portrait into a 512-element L2-normalised identity descriptor, which is then ranked against pre-indexed database entries via Euclidean proximity. Firebase Fire store manages the shared asset catalogue, while a Fast API micro service orchestrates inference and delivery. Experimental trials demonstrate a marked reduction in investigation turnaround alongside retrieval accuracy that surpasses traditional sketch-matching approaches. Keywords - forensic Sketch Recognition, Convolutional Neural Network, Deep Learning, FaceNet512, Feature Extraction, Cloud Computing, Fast API, React, Firebase.