<p>Anomaly detection of multimedia data in fields such as medical imaging, industrial products, and network communication can find abnormal conditions and behaviors, timely identify potential problems, and is of great significance. The feature dimensions in data are the key to the anomaly detection, and the unrelated feature dimensions have a great negative impact on the results of the anomaly detection. Aiming at the problems of hidden outliers which are difficult to handle and the redundancy of feature dimensions in anomaly detection algorithm, this paper proposes a high contrast subspace anomaly detection algorithm based on mutual information related feature dimensions screening (FS-HiCS). Firstly, the algorithm selects the related feature dimensions through mutual information, deletes redundant features. Secondly, the subspace anomaly detection algorithm and high contrast subspace are fused to improve the accuracy of subspace selection and anomaly detection algorithm, and the ability to deal with uneven data. Finally, experiments are carried out on real datasets and synthetic datasets in UCI machine learning database, and the FS-HiCS algorithm proposed in this paper is compared with local outlier factor algorithm (LOF), axis parallel subspace anomaly detection algorithm (SOD), high contrast subspace anomaly detection algorithm (HiCS) and axis parallel subspace anomaly detection algorithm based on related feature dimensions screening (FS-SOD). Experimental results show that the proposed FS-HiCS algorithm improves the accuracy of the anomaly detection and reduces the time complexity.</p>