Temporal Knowledge Graph Completion (TKGC) aims to predict missing facts in knowledge graphs where relations evolve over time. A key challenge lies in effectively modeling heterogeneous temporal dynamics, as some relations exhibit long-term stability while others change rapidly. In this paper, we propose MTF-KGC, a novel time-domain framework that models temporal relation evolution through multi-scale temporal filtering. By treating relation embeddings as temporal signals, MTF-KGC employs a learned temporal filter bank with adaptive gating to decompose relational dynamics into long-term and short-term components. This decomposition enables the model to capture both stable trends and transient temporal patterns in a principled and computationally efficient manner. Extensive experiments on standard temporal knowledge graph benchmarks demonstrate that MTF-KGC consistently outperforms or matches state-of-the-art methods, particularly on datasets with complex temporal dynamics, validating the effectiveness of multi-scale temporal modeling for TKGC.