The images taken by medical capsule with complex noise are difficult to extract features in special stomach environment. In this work, a feature extraction and matching method for the images of stomach environment are proposed. The bilateral filtering algorithm and Multi-Scale Retinex with Color Restoration (MSRCR) algorithm are used to filter and enhance the images, respectively. Furthermore, an improved Scale-invariant feature transform (SIFT) algorithm for gastric image feature extraction is proposed to shorten the time of feature detection. A fast nearest neighbor search algorithm is presented to match the feature points, and the local feature topology constraint method is introduced to match the image features. Finally, experiments are carried out to verify the feasibility of the proposed method. The results show that the feature detection time is reduced by about 16%, the correct matching rate of the fine matching strategy based on local feature topological constraints is about 86%.