Microsoft Word Avtoreferat (Muhiddinov M. N. Tatu)

Figure 1. The proposed salient objects detection and extraction methods 
It calculates an integral image in the first move through the saliency map 
grayscale image and then second move figure out the local window (
SxS
) using an 
integral image for every pixel in constant time further obtain local adaptive three 
level threshold values by performing the comparison of further obtaining local 

---

28 
adaptive three level threshold values. These three thresholds marked the saliency 
map with four different regions, which demonstrates four types of seeds that are 
assigned an iterative variant of the GrabCuts method, allude to as Saliency cuts and 
a full resolution, a high quality binary mask is produced. 
The purpose of proposing a locally adaptive triple thresholding method 
accurately extracts salient objects that have small saliency values. However, due to 
the fact that local thresholding methods are slower than global thresholding 
methods, it has been able to increase the processing speed by computing integral 
images. The integral image (also known as a summed-area table) is a method that 
can be used as a fast and effective way that speed up for computing the sum of 
values (pixel intensity values) in following formulation: 
( , ) = ( , ) + ( − 1, ) + ( , − 1) − ( − 1, − 1)
(3) 
Moreover, the contour of the salient object can be obtained using the binary 
mask and applying Canny edge detection method. These outer boundaries can 
translate into tactile graphics to deliver given natural scene image visual 
information to the visually impaired. In some cases, however, the visually impaired 
may not be confident about the object information just by feeling the boundary of a 
salient object. Therefore, it is very important that internal edges also detected for 
visually impaired to fully understand the salient object in a natural scene. For 
obtaining internal edges of salient object with a help of our saliency cuts’ results, 
which is binary mask, create a matrix with the same size and type as an input 
image to achieve a desired output image. After that, just copy the non-zero 
elements of binary mask that indicate the original input image matrix elements: 
=
( , ).∗ ( , )
(4) 
where 
is the salient object, 
( , )
is the binary mask, and 
( , )
is the input 
image.
Based on the results of the research, salient objects extraction and its contour 
detection in images to create a simple and easy-to-understand tactile graphics has 
been proven to be one of the most effective methods. 
In the third chapter of the dissertation,