Crime scene investigation (CSI) image retrieval is an important means to obtain material evidence for case solving. This paper describes the CSI image datasets

which are classified into different categories according to the content of the data

such as shoe marks

finger prints

tattoo

etc. This paper provides a survey on state-of-the-art techniques in CSI image retrieval focusing on low-level feature extraction and high-level semantic learning. Low-level CSI image features mainly include color feature

texture feature

boundary descriptor

etc. And

three categories of high-level semantic extraction techniques for CSI images are identified including using semantic template and database ontology

machine learning techniques and introducing relevance feedback. In addition

based on practical requirements from the police on using CSI images to find evidence clues

a few research directions are suggested such as introducing prior knowledge of the police to enhance retrieval efficiency.