A improved just noticeable distortion (JND) model based on saliency map detection algorithm is proposed for video coding in order to apply feature of visual perception to the video compression system. The optimal JND model is calculated by setting modle of pixel-based JND with the hybrid. The saliency map concluded by context aware saliency detection modle is used for the weight distribution of JND model. The proposed model and residual filter can be integrated into the framework of HEVC, which is useful for quantifying video data. The experimental results shows that in the case of visual subjective perceptual quality, the average bitrate reduction is 10.71%, compared with HM 16 (all Intra profile).