Large scale medical image search via unsupervised PCA hashing

Medical image search is a significant way to provide similar clinical cases for doctors. Text based and content based image retrieval techniques have been widely investigated in the last decades. However, handling text-missing images and large scale medical database is still challenging. Traditional methods may encounter unsolvable efficiency problem or storage problem when tackling millions of images with general computers. In this paper, we employ an efficient PCA hashing based method for mapping raw features into locality preserving binary code. We focus on investigating the efficiency of PCA hashing while maintaining its competitive performance in medical image search. Ranking aggregation is used to achieve fusion of different features or fusion of retrieval results, which significantly improves single feature retrieval rate and thus compensates the overall accuracy. Without significantly sacrificing the retrieval accuracy, the benefit is a huge gain in physical memory and runtime efficiency. Experimental results show that hashing methods achieve far lower memory and far less time consuming handling large scale database.