Keypoint based moment invariants descriptor for ground-based cloud image retrieval

How to retrieve cloud images from a large cloud image collection becomes an emergent and challenging problem in meteorological area because of the fast accumulation of digital cloud images and the need of cloud automatic observation. This paper aims to address the problem of cloud image retrieval (CIR), which will promote the intelligence of sky imager instruments and help the researchers of meteorology to index and retrieve cloud image. We put forward the keypoint based moment invariants (KeBaMI) descriptor in the framework of CIR. KeBaMI depicts the cloud shape feature with statistical moment invariants based on keypoint, rather than on boundary in traditional approach. Furthermore, we implement the prototype of CIR with KeBaMI. Our experiment results show that KeBaMI is significantly superior over traditional edge based moment invariants descriptor.