This　paper1　proposes　a　mixture　model　that　combines　joint　appearance-based　and　inter-joint　spatial　relationship-based　models　with　a　deep　neural　architecture　called　deep　convolutional　neural　network　(DCNN).　This　method　has　been　applied　to　tackle　the　human　pose　estimation　problem.　Firstly　we　construct　a　graphical　model　for　the　human　body.　Secondly,　the　images　are　decomposed　into　several　image　patches　which　are　used　as　positive　input　samples.　And　finally,　a　DCNN　network　that　can　solve　multiple　classifications　is　obtained　to　perform　human　pose　estimation.　The　quantitative　results　obtained　from　the　experiments　are　impressive　and　show　that　our　method　outperforms　recent　works　of　the　state-of-the-art　that　used　same　considered　datasets.　©　2018　Association　for　Computing　Machinery.　ACM.