Content-Based Image Retrieval (CBIR)

The biggest problem in creating image search systems based on image queries is the semantic gap. A semantic gap is defined as "a lack of agreement between information that people can extract from visual data and interpretation of the same data for users under certain circumstances" (Smeulders). This system can distinguish what is potentially completely different from what the human user can identify from the image. Also, two users may assign different descriptions to the same image, which may complicate the problem.

The other is inverse image search. This is a content-based image retrieval (CBIR) query technology that includes providing a sample image of the CBIR system and searching based on it. In particular, inverse image search is characterized by no search words. This eliminates the need for the user to guess keywords or terms that may or may not return correct results. Inverse image search also allows users to discover the content related to a particular sample image, the popularity of the image, the operation version and the derivative work.

Companies such as DermEngine developed sophisticated dermatological software to automatically identify marker images similar to patient cases using AI algorithm and content-based image retrieval (CBIR). This technique can take up to one year to this process, which helps patients reduce the time to receive appropriate feedback to visit the clinic and grasp their condition. Netherlands-based SkinVision also made significant progress in mobile applications that use machine vision to examine skin lesions and determine the risk of cancer by photographic analysis. This app is trained using a deep learning model to examine over 1 million skin lesion images to help identify unique features such as size, color and shape.

Different CBIR systems, different types of user queries. Typically, tools for content-based image retrieval include query statements and result representations that can be executed by providing a sample image of the sketch or by selecting the color required for the image . Results are displayed by the first few similar images based on similarity. Despite the large number of CBIR systems being developed, there are still many difficult problems in this area. An important aspect that still needs improvement is search speed, accuracy and effectiveness of search results when dealing with large databases. Therefore, researchers in multiple fields are deeply concerned about these aspects.