Object recognition in compressed imagery

Image-based applications can save time and space by operating on compressed data. The problem is that most mid- and high-level image operations, such as object recognition, are formulated as sequences of operations in the image domain. Such methods need direct access to pixel information as a starting point, but the pixel information in a compressed image stream is not immediately accessible. In this paper we show how to perform object recognition directly on compressed images (JPEG) and index frames from video streams (MPEG I-frames) without recovering explicit pixel information. The approach uses eigenvectors constructed from compressed image data. Our performance results show that a five-fold speedup can be gained by using compressed data.