Deep forgery detection technologies play a critical role in image and video recognition. Their performance heavily relies on the features extracted from both real and fake images. However, most existing methods focus primarily on spatial domain features, resulting in a limitation in accuracy. To overcome this challenge, we propose an adaptive region dynamic convolution module that effectively extracts facial features from the spatial domain. Then, an adaptive frequency dynamic filter is proposed to extract the effective frequency domain features. With the fusion of both the spatial domain features and the frequency domain features, the accuracy of classifying real and fake facial images is expected to increase significantly. Finally, the experimental results on three real-world datasets demonstrate the effectiveness of our dual-domain feature representation method, which significantly enhances classification precision.