Global illumination rendering in real-time for high-fidelity graphics remains one of the biggest challenges for computer graphics in the foreseeable future. Recent work has shown that significant amounts of time can be saved by selectively rendering in high quality only those parts of the image that are considered perceptually more important. Regions of the final rendering that are deemed more perceptually important can be identified through lower quality, but rapid, rasterisation rendering. By exploiting this prior knowledge of the scene and taking advantage of image space based algorithms to concentrate rendering on the more salient areas higher performance rendering may be achieved. In this paper, we present a selective rendering framework based on ray tracing for global illumination which uses a rapid image preview of the scene to identify important image regions, structures these regions and uses this knowledge to direct a fraction of the rays traditionally shot. The undersampled image is then reconstructed using algorithms from image processing. We demonstrate that while this approach is able to significantly reduce the amount of computation it still maintains a high perceptual image quality.