License plate recognition systems are used for many purposes, such as for inspecting the behaviour of drivers or locating of stolen vehicles. The development of such systems requires the integration of many approaches of Computer Vision. The proposed method, which intends to detect the license plate on an image, is inspired by the behaviour of human visual system. The primary cortex, which perform frequency analysis, is simulated in order to segment the image. A bank of directional filters in frequency domain is used to extract features from a high dimensional space, and the k-means is used to group them.