Image retrieval using compact deep semantic correlation descriptors

Bo-Jian Zhang, Guanghai Liu, Zuoyong Li, Shuxiang Song

Abstract: Highlights•We propose a channel semantic association learning method for locating potential target objects, which enables target localization without any labelled boxes and category labels.•We propose a hierarchical attention mechanism to integrate multi-level features and distinguish feature importance.•We propose a generalized adaptive PCA-whitening (APW) method to reduce vector dimensionality and improve the retrieval performance of using short vector feature, which is unavailable in existing work.