Introduction: Globally, the rate of crime has dramatically climbed in recent years. Because of the innovative tools used by criminals, controlling crime investigations is difficult. However, a variety of research projects are being carried out in the fields of artificial intelligence and neural networks to automate crime detection and prediction, which has led to the network's strange behavior and necessitated unprecedented finetuning, hyperparameter optimization, and large datasets. For the Multimodal Biometric Crime Control System, a hybridized Convolution Neural Network-Genetic Algorithm (CNN-GA) model was developed in light of the aforementioned information.
Materials and Methods: Facial images and Thumbprint patterns used for the developed system were acquired from publicly available Face and Gesture Recognition Research Network (FG-Net) and Sokoto Coventry Fingerprint Dataset (SOCOFing) respectively. These images were preprocessed using histogram equalization technique to obtain uniform illumination and geometrical size. Procedurally, CNN and GA were used to extract facial and thumbprint features. The extracted features were fused into a single feature set using sum rule strategy. Based on the single feature set, faces and thumbprint pattern were recognized and classified into various individuals using support vector machine (SVM) classifier. The developed CNN-GA was evaluated using computational time (CT) and recognition accuracy (RA).
Results: The result of CNN-GA on fused face and fingerprint at optimum threshold yielded RA and CT of 97.81% and 455.54s, respectively, while the corresponding values of CNN were 95.61%, and 565.02s, respectively. Also, the corresponding values of GA were 96.49% and 560.28s, respectively.
Conclusion: The developed Convolution Neural Network-Genetic Algorithm technique serves as improvement over CNN and GA in terms of recognition accuracy and computational time. This technique could be integrated into emerging crime control systems towards their improved performance.