Abstract: Current point-based rendering algorithms are confined to using Gaussian filter for reconstructing the resulted images. Moreover, some holes may exist in the resulted images from certain viewing directions. In our new approach, an inverse perspective transormation is applied to each point in a 3D model to accurately determine its shape in the 2D screen space. Furthermore, the weight for each fragment is also computed according to this inverse perspective transformation. As a result, our algorithm not only can avoid generating holes in the resulted image, but also can exploit different reconstruction filters in a unified framework. Especially, the resulted image may achieve much higher rendering quality by using conical filter rather than Gaussian filter.