The continuous development of deep forgery technology has led to the proliferation of image tampering and forgery, which contributes to the spread of false information. In order to achieve the detection and recognition of forged images, a forged image feature recognition method and an image tampering detection model are proposed. Among them, the forged image feature recognition method transforms the image into YCbCr colour space and performs DCT transform and LBP transform on the divided image blocks to achieve the description of edge texture and feature learning. The image tampering detection model with improved Faster R-CNN, on the other hand, uses deep learning techniques to learn the features of the original tampered image and introduces an attention mechanism to optimize the classical network. The forged image feature recognition method achieves a reduction in training time consumption (319.9s) while maintaining higher accuracy in feature recognition. The image tampering monitoring model achieves a localisation accuracy (F1) of 0.835 and 0.818 for the two datasets, respectively. Compared with the initial two-stream Faster R-CNN algorithm, the accuracy of splicing detection has improved by 2.7%, copy-movement detection by 2.7%, and tampering detection by removal has improved by 3%, and it has better robustness.