A Low-Dimensional Global Feature Extraction Method Using Composition and Color Information for Similar Image Retrieval

Abstract

In this work, we present a new global feature extraction method by using the composition and color information of images for web-scale similar image retrieval. The proposed feature is robust to transformation such as rotation and it is compressed to low dimensions by using Principal Component Analysis (PCA) for more efficient retrieval. Experiments we conducted illustrate that it is more robust and efficient than the GIST, a state-of-the-art global descriptor commonly used in similar image retrieval.