Just　Noticeable　Difference　(JND)　model　in　transform　domain　is　determined　by　the　contrast　sensitivity　function,　luminance　masking　and　contrast　masking.　In　this　paper,　we　propose　an　improved　JND　model　with　a　new　method　for　contrast　masking　factor　estimation.　We　decompose　an　image　into　structural　image　and　textural　image,　and　the　textural　image　is　used　for　an　accurate　block　classification.　The　proposed　algorithm　can　remove　the　interference　from　the　edge　pixels　thus　the　accurate　JND　in　DCT　domain　is　obtained.　Experimental　results　show　that　the　proposed　algorithm　can　improve　the　JND　thresholds　and　reduce　more　data　redundancy　than　the　relevant　existing　JND　estimators.　This　model　can　be　widely　used　in　many　image/video　processing　fields,　such　as　perceptual　video　coding.　©　2013　IEEE.