Quality-guided image classification toward information management applications

Abstract

Image information management (IIM) is a key technique to improve the performance of large-scale image retrieval. However, IIM is still a big challenge due to the large sum of image datasets and traditional algorithms cannot cope with this problem. In order to solve these disadvantages, we propose a novel image classification algorithm based on image quality assessment (IQA) for image information management. Specifically, we first incorporate both low-level, high-level features as well as quality scores for image representation, where we leverage convolution neural network for deep feature extraction. Then, deep feature vector can be generated by column-wise stacking. Thus, each image can be represented by a feature vector. We leverage GMM to learn the distribution of obtained feature vectors. Similar image categories have similar probability distributions, we leverage the learned GMM model to calculate the posterior probability and image can be classified into corresponding category. Experimental results demonstrate the performance of our proposed method, and image information management is easier to implement.