In recent years, the distribution characteristic of pixel intensity-pairs is proposed, which effectively absorbing the advantages of global and local image enhancement algorithms, avoiding over enhancement, noise amplification and other defects. However, the performance of the algorithm is heavily dependent on the selection of local artificial thresholds. In this paper, by the Just Noticeable Difference(JND) curve from human visual system, dynamic thresholds with visual sense are generated. Then image pixels are divided into different channels for statistical processing, in order to construct mapping functions. Finally, the enhanced image is a weighted fusion of different channels. The experiment results show that the proposed algorithm has better effect and adaptability.