Editorial board editor-in-chief

Implementation of the models
For all ive models the conspicuity maps were created. For the 
models SUN ICA, Bruce & Tsotos and Itti, Koch & Niebur func-
tions and scripts in Matlab environment, provided by the algo-
rithms’ authors, were used to create the conspicuity maps. It is 
worth to mention that the Saliency Toolbox [6] of the Itti, Koch & 
Niebur method [1] presents the inest functionality and is being 
constantly developed. The rest of models – SUN DOG and Zhai 
& Shah – were implemented in Matlab on the basis of authors’ 
publications [3][7]. Sample saliency maps for each model are 
presented in Figure 1.
Fig. 1.
Sample saliency maps for ive attention models: a) original image, b) Zhai & Shah, 
c) SUN ICA, d) SUN DOG, d) Bruce & Tsotos, e) Itti, Koch & Niebur

---

T
elema
tics
52
Effectiveness analysis of selected attention models
Models’ veriication
Subjects:
20 persons (10 women, 10 men) took part in the experiment. 
The subjects age was in range 16 to 50 years.
Experiment
In the study a modiication of the cuing task developed by Posner 
[4] was proposed to verify the attention models. An image, which 
acted as a cue, was presented to the subject on laptop screen. 
There was no precise task as to create condition of free viewing. 
After a deined time (SOA – stimulus onset asynchrony) in one 
of nine possible locations in the image, the stimulus appeared. 
The stimulus was a red dot on a black background. The possible 
locations of dots in the image are shown in Figure 2. When the 
subject noticed the dot, he should have reacted immediately by 
pressing a button on the computer keyboard. A shorter reaction 
time was assumed when the stimulus appeared in a location 
close to the actual focus of attention. Similarly, a longer reaction 
time was assumed in the situation when a stimuli appeared in 
the area where attention was not focused currently.
Fig. 2.
Possible locations of the stimulus
The proposed research methodology required making several 
assumptions. The image was divided into nine, equal rectangle 
areas; the stimulus could appear in the center of any of the nine 
regions. The SOA times were set as 75ms, 200ms and 600ms. 
If the dot appeared in the place where the subject’s attention 
was focused, the response time should have been shortened 
for the SOA = 75 ms and 200 ms and lengthened for SOA = 600 
ms. The assumption of a longer reaction time in last condition 
was connected with the phenomenon of inhibition of return. The 
interval between consecutive images was varied (1000 ms, 1500 
ms or 2000ms) to prevent a development of automatic reaction 
to stimuli. Between the images, a white cross in the center of 
a black background was displayed as a point of concentration 
at which the subjects had to look between images’ exposures. 
A scheme of the experiment is presented in Figure 3.
Fig. 3.
Scheme of the preformed experiment
For each subject a separate experiment was randomly gen-
erated. It consisted of 50 training and 250 test photographs of 
natural, nonemotional scenes taken by authors. The images 
were not framed and showed no particular objects. This was 
done because of the suggestion presented in [3] that framed 
photographs usually have the highest saliency concentrated 
in the center. 
The aim of the training session was to enable subjects to 
become acquainted with the methodology of the experiment. 
The results obtained in these trials were not taken into account 
in the inal analysis. The proper session was divided into two