Analysing texture, color and spatial features for face spoof detection with hybrid classification model

Abstract

Facial biometrics are more natural, innate, and less invasive to humans. It is frequently used in the authentication of approved users and employees to safeguard data from unauthorized access. A face spoofing assault typically consists of a trespasser attempting to impersonate an individual with desirable authentication authorization to gain unlawful access to valuable undisclosed information. In order to detect such violations, several detectives have focused their efforts on visual recognition of structures produced when there are primary signs of spoofing violations. AI techniques like ML/DL play a major role in this aspect. This work intends to propose novel face spoof detection via a hybrid classification model (FSDHC). The input video is initially partitioned into a number of frames and is subjected to the pre processing step to remove the unwanted noise and blurriness from the image with the aid of the wiener filtering technique. Considering the preprocessed video frames, the colour features, improved shape local binary texture, GLCM and HOG-based features are extracted. Subsequently, the extracted features are given as the input to the proposed hybrid classification model to speed up the training process. The proposed hybrid classification model involves two models improved CNN and Bi-GRU models. The final classification process is determined by fusing the intermediate results obtained from both classifiers via an improved score-level fusion process. This work also validates the performance of the proposed model via cross-database validation, in which training and testing with two different datasets like the Replay Attack dataset and MSU MFSD. Additionally, the efficacy of the developed method is assessed using various performance metrics in comparison to state of- art methods.