ISSN 2224-087X (Print)
ISSN 2224-0888 (Online)

Collected scientific papers
"Electronics and information technologies"

(In 1966-2010 published under the title "Electrical engineering")

(Certificate of State Registration 17618-6468 from February 11, 2011)

Main page Search Editorial Policies Rules for Authors

Issue 8

Issue 8, Pages: 93-101
DOI: https://doi.org/
Spectral approach for the template segmentation and object search
V. Parubochyi, R. Shuwar
The objects search and recognition are one of the most common tasks of digital image processing. Since they require high costs of computing, they are the slowest and most resource-consuming parts in a large number of image processing and recognition systems. It creates the need for simple and effective image recognition methods. Unfortunately, the most of the simple methods are also less effective. On the other hand, not all systems require extremely high accuracy and they can use more simple methods with the same success as complex ones.
The article deals with the spectral approach to the implementation of template segmentation and the object search based on a two-dimensional discrete Fourier transform. The spectral template segmentation is a simple modification of the spatial template segmentation which uses Fourier spectrums of the template image and blocks of the analyzed image in the same way as the template image and blocks of the analyzed image are used in the spatial template segmentation. The implemented methods of the spectral template segmentation and object search were compared with the original methods of the spatial template segmentation and object search. On the basis of this comparison, the advantages and disadvantages of the proposed methods were investigated; the speed and accuracy of recognition and search of objects were estimated. Also, the influence of image spatial and texture modifications on the quality of the recognition and object search was considered. For spatial modifications of the image the image rotation at 45, 90 and 180 degrees was considered, and for texture modifications of the image the noise was added, the brightness was increased and decreased.
As a result, conclusions about the possibility of using the proposed methods in practical systems were drawn. Also, a number of approaches to optimize and improve the accuracy of the object recognition based on these methods were proposed.
PDF Version

Main page Search Editorial Policies Rules for Authors

© Ivan Franko National University of Lviv, 2011

Developed and supported - Laboratory of high performance computing systems