<p>Face recognition systems have fostered a high rate of presentation attacks which have left a considerable gap to be filled by the establishment of sound liveness detection techniques capable of functioning in the real world environment. Current methods which rely primarily on a static texture perception or a one-frame analysis usually cannot survive under conditions of high quality replay video or printed facial artefacts. In order to address them, the presented work proposes a dual-stream liveness detection system called BlinkMorph-CADNet that makes the distinction between authentic facial captures and spoofing attempts based on both behavioral and chrominance features. The former stream records the morphology of blinks, such as the change in blink duration, frequency of partial eye closing, and one-sided eyelid movement, which are the involuntary behavior aspects of humans that are not easy to reproduce in replay attacks. The second stream derives spatial chrominance distortion attributes of YCbCr and HSV color spaces with the aim of detecting visual inconsistencies that are brought about by recapturing devices. A consistency-driven decision module is used to combine the two streams in order to build up classification confidence. The experiments of the Anti-Spoofing Replay Dataset publicly available indicate that the proposed model has 97.12% and 95.36% training and testing accuracy, respectively, and 95.82%, 95.14%, 95.48% and 96.07% precision, recall, F1-score, and specificity upon test samples, respectively. The APCER of 1.18% and BPCER of 2.04% obtained are confirmation of the ability to reject attacks. The comparative analysis to CNN, CNN-LSTM, MobileNet and ConvNeXt models show that the models are consistently improving their performance, which confirms the usefulness of the given framework in terms of creating a feasible face liveness detector system.</p>