Research on Intelligent Methods for Intra-Prediction Mode Selection in Video Coding Based on Frequency Domain

With the development of video coding technology, efficient intra-prediction mode selection is crucial for improving compression efficiency. However, traditional methods suffer from high computational complexity due to complex exhaustive searches, making them unsuitable for real-time applications. To accelerate this process, this paper proposes an intelligent method for intra-prediction mode selection based on the frequency domain. By leveraging deep multi-task learning, the method directly predicts the optimal intra-prediction mode for coding units (CUs). The approach first constructs a large-scale frequency-domain database and then designs a multi-task network model capable of simultaneously predicting Matrix-based Intra Prediction (MIP) and Intra Sub-Partitions (ISP) modes. Experiments show that the proposed method achieves a prediction accuracy of over 85%, reduces encoding time by an average of 33.55%, and outperforms traditional spatial-domain acceleration algorithms.