This thesis proposes an automatic face recognition system which is based on the extensions of Local Binary Patterns transformation. The system is composed of three stages: face detection, eye localization - based face alignment, face identification. Face detection module determines the presence of the face in the input image and returns an approximate position and dimension of the subject. Detection of the face provides a rough information about the parameters of the object of interest, thus a second step namely face alignment is incorporated. At this stage dimensions of the facial region are determined more accurately based on the locations of facial features, which in our case are eye pupils. The final module is face recognition which operates in the identification mode and determines the identity of an individual out of a pool of people or rejects an identification attempt. The contributions of the thesis are briefly summarized here: 1) a novel object (face, eye) detection principle , which is based on the combination of Local Binary Pattern histograms with simple classifiers, such as Artificial Neural Network or Support Vector Machine, 2) an accurate face identification algorithm, which is composed of various preprocessing steps, modified Multi-Scale Local Binary Pattern histograms and Weighted Nearest Neighbor Classifier (WNNC), 3) effective mini-batch discriminative feature weighting algorithm supplements the WNNC-based recognition process with statistical data about the classes, which is obtained in the learning process, 4) a fully automatic face recognition system is implemented on TMS320C6416 DSK development board. The first part of the thesis is dedicated to the problem of frontal face detection. A novel face detector which is based on the combination of Local Binary Patterns with ANN or SVM is introduced. The advantage of this setup is the flexibility of the algorithm, which allows to adjust the trade-off between the dimensionality of the feature space and the complexity of the classifier. As the result, the performance which is comparable to state-of-the-art algorithms is obtained in low-dimensional feature space and with simple classifier. The problem of eye localization is covered in the second part of the thesis. The above mentioned principle is utilized in the localization of eye regions in the input face image. For further gain in the localization precision the algorithm is supplemented with the second stage, namely detection of eye pupils. The experiments clearly show that the proposed method outperforms many state-of-the-art eye localization approaches. In the third part the problem of face recognition is addressed. Our face recognition approach is based on the combination of various preprocessing steps, modified Multi-Scale Local Binary Pattern histograms and Weighted Nearest Neighbor Classifier. The key contribution is obtained in the process of weights learning for the Weighted Nearest Neighbor Classifier. The proposed discriminative feature weighting algorithm is robust, fast, requires only \textit{two} training examples per class and can be applied in any multi-class classification tasks. The details about DSP-based implementation of automatic face recognition algorithm concludes the main body of the thesis.