A Study on the Use of Semi-Automatic Systems for Counting Objects in Digital Images

This paper presents an investigation on the impact of coupling a manual correction module to automatic methods for counting objects in digital images, effectively turning them into semi-automatic systems. This study aims to show that completely automatic counting methods often cannot achieve enough accuracy for certain applications and under certain conditions, and that in many cases the introduction of manual corrections is advantageous, despite the impact of this on the time spent by the whole counting process. The paper presents a thorough comparison between fully automatic, fully manual and semiautomatic approaches, using images containing a variety of objects and characteristics, so those approaches could be tested under a wide range of conditions, from the simpler one to the most challenging. It is shown that, with the exception of a few specific situations, the semi-automatic approach is advantageous over the manual counting in terms of time, and over the automatic counting in terms of accuracy. The situations in which the automatic or manual approaches should be preferred are also