Based　on　GPU　parallel　technology,　this　paper　proposes　a　parallel　SRM　feature　extraction　algorithm　to　accelerate　the　extraction　of　SRM　feature　for　steganalysis　of　HUGO　images.　Using　the　parallel　program　framework　of　OpenCL　for　GPU,　we　parallelize　and　implement　a　serial　algorithm　and　employ　some　optimization　technologies　for　our　parallel　program　to　accelerate　the　extraction　process.　The　techniques　include　convolution　unrolling,　combined　memory　access,　aversion　of　bank　conflicts.　The　experimental　results　show　that　the　speed　of　the　proposed　parallel　extraction　algorithm　for　different　size　images　is　25∼55　times　faster　than　the　original　serial　algorithm,　and　2∼4.2　times　faster　than　running　the　parallel　method　on　Quad-core　CPU.　©　2014　IEEE.