In wireless communications, sensitive information is frequently exchanged, requiring remote authentication. Remote authentication involves the submission of encrypted information, along with visual and audio cues (facial images/videos, human voice, and so on). Nevertheless, Trojan horse and other attacks can cause serious problems, especially in the cases of remote examinations (in remote studying) or interviewing (for personnel hiring). This paper proposes a robust authentication mechanism based on semantic segmentation, chaotic encryption, and data hiding. Assuming that user X wants to be remotely authenticated, initially X's video object (VO) is automatically segmented, using a head-and-body detector. Next, one of X's biometric signals is encrypted by a chaotic cipher. Afterwards, the encrypted signal is inserted to the most significant wavelet coefficients of the VO, using its qualified significant wavelet trees (QSWTs). QSWTs provide both invisibility and significant resistance against lossy transmission and compression, conditions that are typical of wireless networks. Finally, the inverse discrete wavelet transform is applied to provide the stego-object. Experimental results regarding: 1) security merits of the proposed encryption scheme; 2) robustness to steganalytic attacks, to various transmission losses and JPEG compression ratios; and 3) bandwidth efficiency measures indicate the promising performance of the proposed biometrics-based authentication scheme.