Image decoding using electroencephalogram (EEG) has became a new topic for brain-computer interface (BCI) studies in recent years. Previous studies often tried to decode EEG signals modulated by a picture of complex object. However, it's still unclear how a simple image with different positions and orientations influence the EEG signals. To this end, this study used a same white bar with eight different spatial patterns as visual stimuli. Convolutional neural network (CNN) combined with long short-term memory (LSTM) was employed to decode the corresponding EEG signals. Four subjects were recruited in this study. As a result, the highest binary classification accuracy could reach 97.2%, 95.7%, 90.2%, and 88.3% for the four subjects, respectively. Almost all subjects could achieve more than 70% for 4-class classification. The results demonstrate basic graphic shapes are decodable from EEG signals, which hold promise for image decoding of EEG-based BCIs.