Abstract

In this paper, we attempt to provide a comprehensive and high-level review of audiovisual features that can be extracted from the standard compressed domains, such as MPEG-1 and MPEG-2. The paper is motivated by the myriad of active research works in extraction and application of compressed-domain features in various fields, such as indexing, filtering, and manipulation. Compressed domain approaches avoid expensive computation and memory requirements involved in decoding and/or re-encoding. Selected features are categorized into four groups -- spatial visual (e.g., color, texture, edge, shape), motion (e.g., motion field, trajectory), audio (e.g., energy, spectral features, pitch), and coding (e.g., bit rate, frame/block type). For each feature, we briefly discuss the extraction methods, computational complexity, potential effectiveness in applications, and possible limitations caused by compress-domain approaches. Finally, we briefly describe audio-visual features specified in the MPEG-7 standard and discuss the possibility of extracting them in the compressed domain.