This paper proposes an adaptive photon tracing approach based on a novel importance function, which combines visual importance and photon path visibility. The generation of photon path is guided by sampling this function to trace more photons to visible and more contributive regions. As a first step, a hierarchy of visual importance maps is constructed. Next, photon paths are produced using a new hybrid mutation strategy, which consists of large mutation and small mutation. The mutation parameter used in small mutation is automatically adjusted using the adaptive Markov chain sampling method. Meanwhile, to find a suitable initial parameter, a mutation parameter initialization method is developed. Experiments show that, compared with previous methods, this approach yields results with better visual quality and smaller numerical error.