The novel extension of the synthetic aperture radar (SAR) technique to the emerging THz spectrum enables a new era of short-range applications such as object recognition, which is presently dominated by the use of optical and infrared sensors. Despite the availability of a large sensing range and penetration depth at the microwave spectrum, analysis of the objects based on geometric shapes is limited due to the lower spatial resolution. In this paper, the detection of geometrical shapes in a high-resolution THz SAR image is investigated. Foremost, a high-resolution environment, which consists of circular, rectangular, and triangular objects is mapped at the frequency spectrum of 325-500 GHz. Further, the map is processed to reduce the artifacts. Finally, the shape detection is addressed based on acquiring the geometric properties of the object. The detected objects can be localized with sub-mm accuracy associated with the spatial resolution.