The work aims to present an algorithm for detecting relevant objects in scanned documents and determining their spatial and logical relationships to classify the document as belonging to one of the specified categories. Graphical objects (e.g. logos, drawings, ornaments, signatures) and alphanumeric objects (e.g. text blocks, tables) will be considered. Based on their relationships (geometric and/or logical), a hierarchical description of the document will be built. Based on the rules created, the document will be classified as belonging to one of the specified categories, e.g., certificate, invoice, official letter, tax return, etc.