Accurate and efficient retrieval of multimedia time series data under uniform scaling and time warping

In this digital age, great interest has been shifted toward multimedia data manipulations. This includes videos, images, and audios, where typical manipulations require fairly large storage and are computationally intensive. Recent research has demonstrated the utilities of time series representation in various data mining tasks, allowing considerable reduction in time and space complexity. Specifically, the utilities of Uniform Scaling (US) and Dynamic Time Warping (DTW) have been shown to be necessary in several human-related domains, where uniform stretching or shrinking, as well as some local variation are typical. Classic examples include a query-by-humnimg system and motion capture data. However, all the past work has neglected the importance of data normalization before distance calculations, and therefore does not guarantee accurate retrievals. In this work, we discuss this concern and present a technique that accurately and efficiently searches under the US with DTW for normalized time series data, where no-false-dismissals are guaranteed.