In order to improve the visual field of photos, an optimized algorithm of automatic image mosaic is presented based on frequency and time domains. The frequency phase correlation is adopted to sort the unorder image sequence and to estimate the overlapping area, and an improved Harris operator is adopted to extract corners. Then, the bidirectional greatest correlative coefficient is used to obtain the initial feature point pairs, and the false feature point pairs are rejected by RANSAC. Finally, the image mean combined with the linear weight function is used to implement image mosaic. Experiment results show that this algorithm can offer a excellent sorting process and show good nomosaic results.The maching speed by proposed algorithm has improved by 40% as compared with those of traditional methods,which efficiently solves the difficulty in confirming corresponding points and achives the desired visual effect in mosaic images with notable illumination difference.