Learning to Understand Image Content: Machine Learning Versus Machine Teaching Alternative

Abstract

Understanding image information content was always a critical issue in every image handling or processing task. Up to now, the need for it was met by human knowledge that a domain expert or a system supervisor have contributed to a given application task. The advent of the Internet has drastically changed this state of affairs. Internet sources of visual information are diffused and dispersed over the whole Web, so the duty of information content evaluation must be relegated now to an image content understanding machine or a computer-based program capable to perform image content evaluation at a distant image location. Development of Content Based Image Retrieval (CBIR) technologies is a natural move in the right direction. However... In this paper the author will argue that the basic assumptions underpinning the majority of CBIR designs are wrong and inappropriate, (like many other basic conceptions that computer vision community proudly holds at this time).