In this year, we modified the image retrieval scheme by using "GAMON" or Videoprint and investigated its application from the viewpoint of building conventional TV broadcasted content retrieval system connected via network

点击翻译按钮获取中文摘要

1. Modification of a feature extraction method to improve its performance : Firstly, in the retrieval method by using "GAMON", we proposed a hierarchical retrieval method, targeted to retrieve the requested video quickly and accurately from the archived video contents in large storage. In the proposed method, one-dimensional DCT operation to a temporal sequence of average color value per frame is executed with quantization of coefficients and moving picture's feature is finally extracted as its lower frequency components. Using this feature as "GAMON", quick and accurate image retrieval becomes possible. As one of experimental results, image retrieval of 34 sequences(TV ads film : each 15 seconds) out of 24-hours-TV content archive, are executed within 3 minutes. Additionally, we evaluated our modified system's performance through previous simulation experiments of image retrieval. As a result, modified system achieved remarkable performance, namely 13 times faster compared to our previous system and 20,000 times faster compared to sequence's time length.2. Application study on conventional TV broadcasted content retrieval system : We assumed a video content distribution system via network with large video content archive. Conventional TV(analog) contents' quality is easily influenced by their receiving environment and even from the same source, different features can be extracted from contents with ghost, for example. Therefore we have intentionally prepared an archive with some contents including ghosts, designed an image retrieval system by using "GAMON" for MPEG-2 images and implemented it. Through simulation experiments, very high accurate image retrieval are done with results of 0 surplus detection and 0 failure detection.