Microsoft Word Avtoreferat (Muhiddinov M. N. Tatu)

«Creation of methods and 
algorithms of salient objects extraction in images for visually impaired»
, 
salient object detection methods based on global contrast enhancement and salient 
object extraction method based on locally adaptive triple thresholding were 
developed. The proposed method consists of the following steps as shown in 
Figure 1: 1) global contrast enhancement method using histogram equalization, 2) 
saliency map generation method 3) salient region extraction method based on 
locally adaptive triple thresholding, and 4) salient object’s boundary and inner 
edge generation method.
Global contrast enhancement based on histogram equalization of color values 
of the image is applied into pre-processing step to improve salient region contrast 
than surrounding regions. A high contrast against the surrounding region is usually 

---

27 
a stronger evidence for saliency of a region than a comparable contrast against a 
far-away region. Naively evaluating the saliency of a pixel is defined using its 
color contrast to all other pixels in the image and takes 
(
)
time, which is 
computationally too expensive even for medium sized images. To reduce 
computational time, contrast enhanced image segmented into regions using graph-
based images segmentation method and regions’ saliency value can be obtained by 
measuring its color contrast to all other regions in the image: 
( ) = ∑
( ) ( , )
(1) 
where 
( ) 
is the weight of region and 
( , )
is the color distance metric 
between two regions. Spatial information is incorporated by using a spatial 
weighting term to increase the effects of closer regions and to decrease the effects 
of farther regions. In the spatial weighting formulation, 
( , )
is the spatial 
distance between regions and , 
control the strength of spatial distance 
weighting, 
( )
is the weight of region defined by the number of pixels in , 
( ) =
(−9 )
is a spatial prior weighting term, and 
is the average 
distance between pixels in region and the center of the image and 
= 0.4
. 
( ) =
( ) ∑
,
( ) ( , )
(2) 
In addition, salient object extraction method based on locally adaptive triple 
thresholding using integral images is proposed. The proposed salient object 
detection and extraction methods consist of pre-processing, saliency map, saliency 
cuts, outer and inner edges as shown in Figure 1.