The deliverables of an airborne LiDAR survey usually include all points, ground points, digital surface models (DSM) and digital elevation models (DEM). DSM and DEM are raster products. DSM is a grid interpolated from all points whereas DEM is a grid interpolated from ground points only. The difference of DSM and DEM is known as a normalized DSM denoted as nDSM which expresses the heights of objects off-the-ground. Shallow landslides are the most common landslides triggered by torrential rainfalls and explicit fresh scars after rainfall events. The pixel value of landslides on nDSM is approaching zero due to its bareness. Therefore, this property can be applied for deriving useful thematic maps for subsequent landslide detection. Thus, point operation of image enhancement is employed, including image subtraction method and gray-level slicing method. Images of various slicing intervals are subsequently segmented on basis of object-oriented approach and processed by morphological filters to obtain vector shapes of landslides. Finally, classification results are verified by the result obtained by manual interpretation of the aerial photographs taken at the same time. The experiment is carried out using a LiDAR dataset in I-Lan County acquired after Typhoon Kalmaegi on 17 July 2008. The overall density of all points is 1.454 points/m2 and ground point density is 0.454 points/m2. Ground check with 60 observations shows an average vertical accuracy of 0.057 m. The results show that (1) DSM and nDSM are better than DEM for landslide interpretation. (2) Cares must be taken to treat the outliers in a nDSM model. (3) Shallow-seated landslides can be effectively enhanced by applying a level slicing to nDSM with lower level bound of 0.3 m and upper level bound of 5 m. (4) OOA and morphological filtering are effective approaches for obtaining solid shapes of landslides in an automatic classification.