| Summary: | 碩士 === 國立中央大學 === 通訊工程研究所 === 100 === With the development of internet and mobile communications, efficient multimedia retrieval from huge databases becomes an important issue. Although content-based image retrieval has been extensively studied in recent years, there still exist shortcomings. In particular, extracting features from fully decompressed image is computation and memory resource consuming.
This thesis focuses on the content-based image retrieval in compression domain. The extracted features are based on the I-frame coding information in H.264 that is a powerful and widely used coding standard. We propose to employ the local mode histogram as the texture feature to match images, and apply the residual coefficients to filter non-confidence modes. Moreover, the geometrical correspondence between two images is also considered in our method.
The experimental results show that the proposed method can reduce the resource consumption and archive the similar performance, i.e., MAP 0.3, and ANMRR 0.625 in Oxford 5k compared with the method that extracting features in decompressed images.
|
|---|
