Dynamic Texture Recognition Based on Compression Artifacts

Abstract

The paper proposes a novel approach to the classification of compressed videos containing dynamic textures. The term dynamic texture is usually used with reference to image sequences of various natural processes that exhibit stochastic dynamics (e.g., water, fire and windblown vegetation). Description and recognition of dynamic textures have attracted growing attention.Although one of the most important prospective applications of the technology is content-based video retrieval, recognition of dynamic textures for compressed video has not been considered. The content of video and dynamic textures in particular, profoundly affect the performance of video compression algorithms. The prominence of compression artifacts can, therefore, be used to recognize dynamic textures in compressed videos. In the paper, we show how features, previously proposed for quality assessment, statistical analysis and a soft computing technique (neural networks) can be used to discern 23 different classes of dynamic textures in a standard video database, with 99.5% accuracy.