With　the　widespread　application　of　image　editing　software,　image　manipulation　localization　has　become　a　focal　point　and　promising　research.　Existing　neural　networks　for　image　manipulation　primarily　rely　on　RGB　and　noise　features　to　accurately　identify　tampered　areas　within　images.　However,　in　practical　image　manipulation　localization　tasks,　noise　features　extracted　from　RGB　images　alone　are　often　insufficient　to　effectively　address　tampering　issues.　Furthermore,　existing　encoder-decoder　models　for　image　manipulated　localization　often　overlook　the　direct　interactions　between　different　layers　during　the　decoding　process,　which　hinders　the　effective　transfer　of　deep　semantic　information　to　shallow　features,　thereby　impacting　the　ability　to　accurately　identify　manipulated　areas.　To　address　the　challenges　previously　identified,　this　paper　presents　a　dynamically　adaptive　noise　extraction　module　and　achieves　inter-layer　information　exchange　in　the　decoder　by　fusing　output　features　from　different　layers　to　extract　edge　information.　We　adaptively　map　RGB　images　to　an　appropriate　color　space　using　linear　transformations　and　then　extract　noise　features,　leveraging　the　differences　in　color　blocks　to　effectively　uncover　features　of　tampering.　In　addition,　we　integrate　features　across　multiple　decoder　layers,　employ　deep　multi-scale　edge　supervision　to　impose　constraints,　and　introduce　a　dynamic　ringed　residual　module　to　further　enhance　feature　representation.　Extensive　experiments　demonstrate　that　our　approach　achieves　competitive　results　on　diverse　large-scale　image　datasets,　exhibiting　superior　precision　and　robustness　compared　with　most　state-of-the-art　methods.