PatMedia is a Hybrid Retrieval Engine for patent multimedia content developed at the Informatics and Telematics Institute. The system is capable of retrieving patent figures by combining visual and textual information. From the functionality point of view, PatMedia supports the submission of queries by visual example, textual search and concept-based retrieval. PatMedia comprises the retrieval module of an integrated patent image retrieval framework that supports automatic extraction and indexing of patent figures. Visual Search is realized with an advanced algorithm for the extraction of an innovative feature for binary image retrieval, named the Adaptive Hierarchical Density Histogram, which represents an image based on its geometry with a view to finding visually similar content. Text Retrieval is based on the textual description of the patent figures in the document, while matching of the text with the images is performed by recognizing figure labels with the aid of OCR techniques. The search options can be combined with filters on figure type and textual description. The concept search functionality of PatMedia is based on of a supervised machine learning framework using Support Vector Machines (SVM). Specifically, an individual SVM classifier is trained for every defined high-level concept to detect the corresponding instances based on visual and/or textual low-level descriptors.

Further information on PatMedia can be found at http://mklab.iti.gr/content/patmedia. 

An evaluation of PatMedia by a patent information professional was presented at the IRF Symposium 2010: Download the presentation.