This paper deals with various techniques used in the context of automatic image analysis. Content is intended to provide an overview of techniques used to extract information from images, with a concrete example. The paper describes how a program extracts information from two photographs of the Rubik's cube to determine the state of the cube. The program's work is divided into distinct steps with each step explained individually. Techniques covered include edge detection, shape classification of polygons, white balancing and comparison of colors.
An important part of the work is based on an understanding of how unstructured data can be processed into structured data. The process makes assumptions and estimates, so the information can be structured and used in the next step. The document shows how information about the analyzed object can be used to detect errors in the assumptions and estimates made. For example, the labels must not have colors that result in physically impossible pieces of the cube. The paper contains a result part where the program is demonstrated for photos of different quality.