A New Way for Face Sketch Construction and Detection Using Deep CNN

Abstract

Traditional hand-drawn face sketches have encountered speed and accuracy issues in the field of forensic science when used in conjunction with contemporary criminal identification technologies. To close this gap, we provide a ground-breaking research article that is built on a stand-alone program that aims to revolutionize the production and identification of composite face sketches. This ground-breaking approach does away with the requirement for forensic artists by enabling users to easily create composite sketches using a drag-and-drop interface. Utilizing the power of deep learning and cloud infrastructure, these generated sketches are seamlessly cross-referenced against an enormous police database to identify suspects quickly and precisely. Our research study offers a dual-pronged approach to combating the rise in criminal activity while using the quick breakthroughs in artificial intelligence. First, we demonstrate how a specific Deep Convolutional Neural Network model transforms sketches of faces into photorealistic photographs. Second, we employ transfer learning for precise suspect identification using the pre-trained VGG-Face model. Utilizing Convolutional Neural Networks, which are famous for their data processing powers and hierarchical feature extraction, is a key component of our strategy. This approach exceeds current methods and boasts an extraordinary average accuracy of 0.98 in identifying people from sketches, providing a crucial tool for strengthening and speeding up forensic investigations. A unique Convolutional Neural Network framework that demonstrates significant improvements over state-of-the-art techniques is also revealed as we dive into the challenging task of matching composite sketches with corresponding digital photos. Our thorough analysis shows the framework to be remarkably accurate, constituting a substantial advance in the field of forensic face sketch production and recognition.