Multi-spectral sensor could record the spectral response of various land cover from visible light to near infrared information. In this study, a Miniature Multiple Camera Array (MiniMCA) mounted on a fixed-wing UAV is adopted. The spatial resolution could reach to 0.26 meters with a flight height about 500 meters. The MiniMCA-12 is composed of twelve cameras, which is a multi-spectral, multi-lens, and frame-based optical sensor. The spectral resolution for each band is 10 nm that covers from 400 nm (blue) to 950 nm (near infrared) spectral range with close regard to known biophysical indices suitable for environmental observation purposes. However, due to each band of the camera is acquired by individual perspective center, it is necessary to perform band-to-band registration before remote sensing applications. In this paper, two methods are proposed for band registration, called "object space" and "image space" methods. Since the short focal length (9.8 mm) of MiniMCA-12 cameras may cause significant lens distortion, the Interior Orientation Parameters (IOPs) of all cameras have to be calibrated in advance. For object space method, the relative orientation parameters (ROPs) were obtained through the exterior orientation parameters (EOPs) during camera calibration. Further, we can perform aerial triangulation on the master camera with ground control points to obtain all master images' absolute EOPs, and the absolute EOPs of the other 11 slave images can be computed through the ROPs. Later, the ortho-rectified images can be generated by using each image's EOPs. On the other hand, for the image space band registration method, the concept of plane-to-plane perspective projection, also called 8-parameters transformation, is adopted. Together with lens distortion correction, all of the slave cameras were rectified into the same image plane of the master camera, thus all 12-band images will have the same IOPs and EOPs as the master camera during ortho-rectification. Experimental results show that one pixel miss-registration can be achieved for the object space method, but four pixels of transformation error were observed for the image space one.