Imagine is a Content Based Image Retrieval (CBIR) application, in other words it’s a search engine for images, able to analyze and search images also based on their content and not only on their textual metadata.
The name Imagine is the contraction of words IMAge search enGINE and consciously recall the english verb imagine, that is the ability to figure images in mind. Imagine is developed in Java and uses, for indexing and search, Elasticsearch a distributed and scalable search engine based on Apache Lucene.
Imagine implements and uses algorithms of Information Retrieval field, adapting them to the image context. Images are analyzed through global and local feature extractors. This allows to provide to the users different search mode.
More in details, Imagine provides three different search mode:
In the video, in order to show all the features just exposed, I used UKBench, a dataset known in the scientific community. The dataset counts 10.200 images grouped in set of 4 elements. Each set contains 4 images of the same object taken by a different perspective. Using this dataset, the better achievable result, for each search, consists in a result set having at the first 4 positions all the 4 variants of the image used as query.