Dataset of Mugshots

The major goal of the self-created dataset is to eliminate the class imbalance in LFW i.e., the dataset's largest class has 500 times more photos than the smallest. As a result of this flaw, the model may bias towards the class with more photos. Another reason for creating the criminal dataset is to account for low-resolution photos, as standard datasets use high-resolution cameras to gather photographs. To show the real-time application of face recognition, a self-created dataset containing 10 classes of criminals is used. Mislabelled and vague photos are manually removed, and 25 images from each class are considered to create a class-balanced dataset. Out of which 10 images of each class are used to construct low-resolution images and partially occluded faces. Therefore, each class of the proposed dataset is having 35 images. Photographs of the criminals for the self-created dataset are taken from Google. As the captured video frames contain numerous faces at a time, another testing dataset is made to include 50 different photos with multiple faces, thus the planned dataset is now ready to demonstrate the real-time surveillance system.