The aim of this paper is to improve the accuracy of copy-move forgery detection (CMFD) in image forensics by proposing a novel scheme. The proposed scheme integrates both block-based and keypoint-based forgery detection methods. Firstly, speed-up robust feature (SURF) descriptor in log-polar space and scale invariant feature transform (SIFT) descriptor are extracted from an entire forged image. Secondly, generalized 2 nearest neighbor (g2NN) is employed to get massive matched pairs. Then, random sample consensus (RANSAC) algorithm is employed to filter out mismatched pairs, thus allowing rough localization of the counterfeit areas. To present more accurately these forgery areas more accurately, we propose an efficient and accurate algorithm, evolving circular domains coverage (ECDC), to cover present them. This algorithm aims to find satisfactory threshold areas by extracting block features from jointly evolving circular domains, which are centered on the matched pairs. Finally, morphological operation is applied to refine the detected forgery areas. The experimental results indicate that the proposed CMFD scheme can achieve better detection performance under various attacks compared with other state-of-the-art CMFD schemes.