Digital image content analysis
Image can be analysed in many ways – from artistic impression, colour and spatial composition, through information content to its function in visual communication. An image can be abstract or technical, greyscale or colour, clear or foggy etc. Our interest is to analyse the content of the digital image, that is to find semi-automatic procedures for at least partial understanding of the meaning of its content. What is the substance of the digital image, what does it consist of? An image can be thought as a sequence of objects disposed on the background of the image. Loosely speaking, an image object is an image region having certain meaning by itself. The aim of the analysis is to understand the meaning of such objects and to analyse relations among them. Nowadays, the analysis of information has become of paramount importance. Image analysis can also be seen as an image information extraction. Every image carries a huge amount of information but only a small part of it is relevant for a certain application.The proverb says “One picture is worth more than thousand words” but it does not say which picture means which thousand words. The goal of image analysis is to find some of those words, that is to identify relevant information, neglect the irrelevant one and, according to the information gathered, do some action.
How to achieve such goals? Our attempts are listed bellow.
Digital images are huge square arrays of numbers and analysing it is actually doing calculations on those numbers using modern computers. It turns out that it is somehow difficult to give a meaning to a whole array of numbers. One way to identify image objects is to mark their background. Therefore, the analysed image is first divided into regions called shapes and those regions are coded in order to make further analysis possible. Usually, an image shape is described by its boundary which is not always the most appropriate way. A preliminary study of shape coding based on whole regions is given here. Certain properties of shapes can be analysed using morphological operators. Here is the introduction.
When certain image regions are identified, how to recognize their identity and relations between them? It turns out that the identity of one object depends on the whole scene brought by the digital image. Even more, the meaning of the digital image does not depend only on the identity of objects but also on the relations among them. How to gather global information when digital image analysis is mostly locally oriented? A formal system was introduced to formalize manipulations with abstract symbols, to describe the creations of symbols, their interaction, recombination and analysis of their meaning. It seems the formal system provides an adequate context to describe image objects, their structure and analyse their meaning. Beside that, it offers valuable hints for programming language data structures construction and architecture. Our attempt of digital image scene analysis and object recognition based on formal system is described here and here.
Not only theory
Digital image analysis can not be only theory - even the most abstract assumptions should be verified in computer simulation. Among other, it seems that the efficiency of development depends on the number of trials (successful and failed) performed in a given interval of time. To be operative, an C++ environment for digital image and video processing was developed in our group.
Why the digital image analysis is so important? Because it allows a communication between a machine and its environment and a man and a machine. It turned out that the user interfaces and not the underlying technology are the crucial parts of nowadays electronic equipment. To be honest, keyboard is not very user-friendly device. How to use my VCR? How to select the right item out of 20.000 hits found by the AltaVista search engine? By user identification, modelling and user interface personalization. A short lecture on user identification is provided here.