In general, biometrics aims is the identification or verification of individual, especially using their physical or behavioral characteristics. This practice tends to replace the traditional knowledge-based methods such us a password or PIN code and token-based methods such as identity document or a badge. Daily, multiple biometric modalities have been developed, where the products are available and already used in many applications. Biometric recognition is a research area that does not stop evolving and seeking new forms of high performance modalities. The main of this thesis is to develop and evaluate new methods based on hidden biometric features, tamper-proof and can't be voluntarily changed. In this context, that we introduce a new biometric modality that using human brain characteristics and the feasibility of such a method was the object of our study. For this, brain volumetric images, obtained by MRI (Magnetic Resonance Imaging) are used to extract the most discriminative brain patterns. Afterward, biometric code of the brain, called « BrainCode », is generated that serve on individual identification or authentication. Thus, we developed three biometric techniques based on the brain. The first technique uses textural patterns of a brain digital image, while the second technique is based on the use of morphological and geometrical characteristics of the brain. The last explored technique, based on the fusion of geometric features and the textural patterns from brain MRI slice. These new biometric techniques obviously require the acquisition of brain MRI images by considering only healthy and adult peoples. According to obtained results from experiments, the developed techniques lead to interesting recognition performance. More precisely, the first technique based on texture patterns analysis and « BrainCode » generation, provides about 97,53% of accuracy, FAR = 1,5%, FRR = 3,41% and the EER = 2,72%. The second technique, using a geometric model of the brain, called « GMB » (Geometric Model of the Brain), we obtained a maximum accuracy around 98,80%, FAR = 0,09%, FRR = 2,31% and the EER = 1,92%. Finally, the merger of geometric features and the texture, we have reached about 99, 47% of accuracy, FAR = 0,32% and the FRR = 0,72%. In this study, we are also interested on the robustness study of the proposed approaches against noise
