In this work, we propose a technique for facial expression recognition to bridge the semantic gap among the features that can be extracted in a content-based video retrieval system. The paper aims to provide accurate and reliable facial expression recognition of a dominant person in video frames using deterministic binary cellular automata (DBCA). Both geometric and appearance-based features are used. Efficient dimension reduction techniques for face detection and recognition are applied. Using the facial action coding system (FACS), one can code automatically nearly any anatomically possible facial expression, deconstructing it into what are called as action units (AUs). By employing two-dimensional deterministic binary cellular automaton systems (2D-DBCA), a scheme is developed to classify the facial expressions representing various emotions to retrieve video scenes/shots. Extensive experiments on Cohn–Kanade database, Yale database, and large movie videos show the superiority of the proposed method, in comparison with support vector machines (SVMs), hidden Markov models (HMMs), and neural network (NN) classifiers.