In this paper, a new algorithm (Image Recognition Volume Rendering ---- IRVR) is developed, which can remarkably reduce the redundant data during volume rendering. The IRVR algorithm applies the cross-entropy threshold selection method to identifying the object and background voxels from a three dimensional dataset. An iterating ray tracing method combined with the objects detection sampling strategy is developed to sample the original dataset. The fast iterative method is applied to sampling the classification dataset for locating the intersection position of the viewing ray and the original dataset. And the precise normal sampling method (for example, trilinear interpolation, spline interpolation, etc.) is used to extract the gray value from the original dataset for volume rendering. The experiment result is given in this paper. It is proved that the IRVR algorithm can increase the volume rendering speed and guarantee the quality of the volume rendering results.