In the law enforcement investigation, the police use sketching techniques to identify suspects from an eyewitness's memory. Many automatic face sketch recognition systems that determine the perpetrator’s appearance from the face image datasets have been proposed. The aim is to conduct the arrest of the right offender. We propose this work to carry out a search based on the ethnicity criterion to speed up this automatic identification and to help authorities execute fast responses by launching the retrieval process only in a part of the dataset of face images. The goal of this study is to enhance the accuracy of ethnic face sketch classification by using the convolutional neural network built on the VGG16 architecture. The FairFace dataset, which includes seven ethnic face images: White, Black, Indian, East Asian, Southeast Asian, Middle Eastern, and Latino|Hispanic, was employed in the study. We convert the face images dataset to face sketch images, and we optimize the VGG16 model for seven classification outputs. This work shows that the VGG16 deep learning model offers a reliable, automated approach for ethnic face sketch classification and recognition. The used model achieved an accuracy reaching above 94% and produced a low false negative rate, which is crucial for minimizing undetected cases.
Ounachad et al. (Thu,) studied this question.