Cloud Classification using K-Means Clustering and Content based Image Retrieval Technique

Presently whole sky imagers (or ground based imagers) are becoming popular because of the upward pointing nature of the imaging system and also its ability to capture the ground truth relevant to images being captured. These imaging systems also provide high resolution images which gives an added advantage. In this paper we present a system which makes use of ground based images of clouds in order to classify the cloud as high level, middle level and low level clouds. We make use of k-means clustering and Content Based Image Retrieval (CBIR) techniques for cloud classification. The system developed classifies the clouds as low level, middle level and high level clouds. Cloud class plays a vital role on deciding the rainfall precipitation. The result of this cloud classification can be later used as an input to a system which dynamically decides rainfall precipitation. For this purpose, identification of the low-level clouds is of high importance.