Face detection is considered to be one of the principal techniques of biometrics. Several methods for face detection have been proposed and described in the literature, but the Viola and Jones method is one of the most prominent. This method is based on the principle of Haar filters. In this study, we propose a new type of Haar filter called a dispersed Haar filter. This new structure provides more flexibility for very complex geometry, such as the human face. To create the structure of the filter, we used three optimizations methods: differential evolution (DE), genetic algorithm (GA), and particle swarm optimization (PSO). To test our approaches rigorously, we performed two types of tests. The first test is facial detection on fixed images from three different databases (Caltech 10K, FDDB, and CMU-MIT), which presents a significant challenge. The second test is more efficient and involves the recognition of human faces from a video database. For our experiment, we used a YouTube celebrity dataset. This system consists of two stages: Face detection using three detectors: Haar-DE, Haar-PSO, and Haar-GA. Face recognition using three machine-learning algorithms: multilayer perceptron (MLP), support vector machine (SVM), and convolutional neural network (CNN) with multi-scale images. The proposed Haar-DE algorithm demonstrates good detection performance on several databases compared with the state-of-the-art methods.