Cloud Image Retrieval Method Base On Topic Model

Cloud is an important part of water and energy cycle in the atmosphere. It demonstrates the extent of stability, moisture conditions, height and thickness, and it is a key feature to predict future weather. In recent years, Sky-imaging system shave achieved considerable development, and generate a large scale of cloud images. It is difficult to browser each cloud image in a large image set. Therefore, ground-based cloud image retrieval systems have become aurgent requirement.With the help of the traditional content-based image retrieval (CBIR) technology and topic models, we research the ground-based cloud image feature representation and their retrieval models.The main research work includes three aspects:This paper presents the concept of atomic cloud and a new cloud image retrieval method base on bag-of-visual-words (BoW) model. Atomic cloud refers to local features extracted from the basic feature cloud imageby a unsupervised learning. The new cloud image retrieval method integrates with atomic cloud, the state-of-the-artsearch engine, BoW architectureand inverted indexingtechnology. The proposed method outperforms the traditional CBIR methods.This paper presents a new cloud retrieval method based on topic models. This method combines with BoW architecture, generating the topic models for the virtual word on the cloud image. By learning the relevancedegree between cloud imagesand latent topics, the similarity of the two cloud images can be evaluated. Out experiment results show that this method achieves better performance than that of BoW method.This paper designs and implements an online platform for digital ground-based cloud image retrieval, including control module, feature extraction, visual words, retrieval methods, website and performance evaluation system.