Detection　of　image　forgery　is　an　important　part　of　digital　forensics　and　has　attracted　a　lot　of　attention　in　the　past　few　years.　Previous　research　has　examined　residual　pattern　noise,　wavelet　transform　and　statistics,　image　pixel　value　histogram　and　other　features　of　images　to　authenticate　the　primordial　nature.　With　the　development　of　neural　network　technologies,　some　effort　has　recently　applied　convolutional　neural　networks　to　detecting　image　forgery　to　achieve　high-level　image　representation.　This　paper　proposes　to　build　a　convolutional　neural　network　different　from　the　related　work　in　which　we　try　to　understand　extracted　features　from　each　convolutional　layer　and　detect　different　types　of　image　tampering　through　automatic　feature　learning.　The　proposed　network　involves　five　convolutional　layers,　two　full-connected　layers　and　a　Softmax　classifier.　Our　experiment　has　utilized　CASIA　v1.0,　a　public　image　set　that　contains　authentic　images　and　splicing　images,　and　its　further　reformed　versions　containing　retouching　images　and　re-compressing　images　as　the　training　data.　Experimental　results　can　clearly　demonstrate　the　effectiveness　and　adaptability　of　the　proposed　network.