Abstract: We propose a novel parallel image space adaptive rendering approach. Contourlet transform which includes Laplacian pyramid and directional filter banks is modified for multi-scale and directional analysis in our algorithm. In sampling stage, the image space is coarsely sampled first. The sampled image is analyzed into coarse and difference values in multi-scale using Laplacian pyramid transform. Based on the analysis, a heuristic method is proposed to repeatedly distribute adaptive Monte Carlo samples. In reconstruction stage, the final image is reconstructed by filtering each pixel using our anisotropic per-pixel filter. The filter size depends on the variance and attenuation values. The filter's anisotropic property is computed by the directional filter banks. Compared to the state-of-the-art image space adaptive rendering methods, the results rendered by our algorithm show improvement in both visual image quality and numerical error while using sparse samples.