With　the　increasing　number　of　fake　images　on　the　Internet,　the　detection　and　localization　of　such　images　have　become　a　topic　worthy　of　attention.　However,　existing　methods　generally　have　the　following　problems:　Single-type　detection　struggles　to　address　the　complexities　of　diverse　real-world　scenarios;　Over-reliance　on　specific　situations　limits　the　practical　effectiveness　of　statistical　methods　in　Image　Manipulation　Localization;　The　backbone　feature　extraction　network　during　training　often　misidentifies　high-contrast　regions　as　manipulated　areas.　In　response　to　these　problems,　this　paper　introduces　a　novel　approach　named　Progressive　Multiscale　Fusion　Network　for　image　manipulation　localization.　To　begin　with,　an　Edge　Trace　Block　is　designed　to　extract　multiscale　edge　features　and　perform　edge　supervision　so　that　PMF-Net　can　obtain　global　context　information　on　edge　parts,　including　trusted　tampering　edge　clues.　Subsequently,　we　propose　an　innovative　approach　named　Attention　Fusion　Block　that　fuses　the　features　of　two　different　sources　using　an　attention　map,　and　then　further　extracts　the　tampering-related　information　with　lightweight　attention.　Extensive　experiments　show　that　our　method　outperforms　state-of-the-art　works　in　both　localization　performance　and　robustness　on　several　benchmark　datasets.　©　2025　IEEE.