A new approach

image registration and superposition of 3D reconstruction based on CT images is proposed in this paper. Compared with the traditional approaches

it is simpler and more effective. It can be realized in three steps. Firstly

correcting the distortion of all the 2D CT images

some images between every two serial CT slices should be interpolated if needed. Secondly

every pixel will be endowed with information of depth and transparence according to its spatial coordinates and gray level. Finally

a 3D image will be displayed on the screen. The reconstruction image includes not only information of the surface of the object

but also the inner of the object. The results of experiments indicate that the 3D image reconstructed has only a little distortion and the results of measure by software are very accurate. The experiment was carried out successfully with 33 pieces of serials image of mandible

a accurate 3D image of mandible was acquired. Compared the results measured by software in our reproduction model with those measured by micrometer in true model

the standard deviation of absolute error is only 0 151mm. At the same time

this registration method can be applied to many other cases for 3D image reconstruction for serial sections.