Block-based　Discrete　Cosine　Transform　(BDCT)　image　compression　method　inevitably　produces　annoying　blocking　artifacts　in　the　case　of　low-bit-rate　compression,　as　each　block　is　transformed　and　quantized　independently.　Blocking　artifacts　not　only　seriously　affect　the　subjective　image　quality,　but　also　affect　the　performance　of　automatic　analysis.　In　this　paper,　we　propose　a　pre-classified　sparse　representation　based　deblocking　method.　We　combine　the　human　visual　sensitivity　based　classification　and　the　sparse　representation　method　together.　For　different　contents,　the　reconstruction　threshold　can　been　adaptively　adjusted.　Experimental　results　show　that　the　proposed　method　can　improve　the　deblocking　quality　of　the　compressed　images　effectively.　©　2015　IEEE.