Mirror visual feedback (MVF) is used widely for motor recovery after stroke, but an optimal training setup and systematic procedure are lacking. New optimization strategies have been proposed, one of which is a camera technique. We investigated the effects of a camera-based MVF setup on motor function and motor processes upstream for upper-limb rehabilitation. Seventy-nine stroke patients were assigned randomly to the MVF group (MG; ${N} = {38}$ ) or conventional group (CG; ${N} = {41}$ ), which, respectively, received camera-based MVF and dosage-equivalent physiotherapy or/and occupational therapy for 1 h/day and five days/week for four weeks. Two clinical scales were used to quantify the effect of the intervention methods: the Fugl-Meyer assessment-upper limb (FMA-UL) subscale and Barthel index. The hand laterality task was used to evaluate the ability of mental rotation, including the reaction time and accuracy. All measurements were improved significantly for both groups following intervention. FMA-UL was improved significantly in the MG compared with that in the CG. In lateralization tasks, the RT of the MG was significantly shorter than that of the CG at the endpoint. For all patients, judgments for the affected side were significantly slower and less accurate than for the less-affected side. Subgroup analyses suggested greater benefits of motor function, the activities of daily life, and mental rotation were achieved in subacute patients after MVF. A trend toward greater improvements in motor function for patients with severe-moderate motor impairment and patients with right-hemisphere damage were also revealed. Camera-based MVF has improved the motor function and ability of mental rotation for stroke patients, especially for patients in the subacute stage, which indicates the potential to improve motor preparation. Further studies might combine mental rotation with electroencephalography to investigate the neuro-mechanism of MVF.