Owing　to　the　irregular　distribution　of　point　clouds　in　3D　space,　effectively　compressing　the　point　cloud　is　still　challenging.　Recently,　numerous　compression　methods　have　been　developed　with　outstanding　performance　for　the　compression　of　geometry　information　of　point　clouds.　On　the　contrary,　limited　explorations　have　been　devoted　to　point　cloud　attribute　compression　(PCAC).　Thus,　this　paper　focuses　on　the　study　of　point　cloud　attribute　compression　by　applying　geometric　information　as　decoded　prior　information.　In　this　paper,　a　novel　cluster-based　two-branch　framework　for　PCAC　is　proposed.　Specifically,　the　point　cloud　is　first　divided　into　adaptive　un-overlapped　blocks　via　K-means　according　to　geometric　information,　which　enables　an　efficient　local　representation　of　the　point　cloud.　Then,　we　split　the　point　cloud　attributes　into　two　components(Block-Mean　component　and　Block-Residual　component)　and　compressed　them　separately　through　the　designed　block-based　dual　branches.　For　the　Block-Mean　component,　we　design　a　prediction　scheme　to　remove　the　inter-block　attribute　redundancy.　The　Block-Residual　component　is　further　compacted　by　applying　a　designed　graph　Fourier　transform　generated　by　geometric　prior　information.　Then,　the　transform　coefficients　are　encoded　using　an　arithmetic　coder.　Extensive　experimental　results　demonstrate　that　the　proposed　point　cloud　attribute　compression　scheme　outperforms　existing　attribute　compression　schemes　in　terms　of　both　objective　quality　and　subjective　quality.