Preču zīmju meklēšana attēlos
Loading...
Date
Authors
Advisor
Journal Title
Journal ISSN
Volume Title
Publisher
Latvijas Universitāte
Language
lav
Abstract
Šajā darbā tiek pētīta preču zīmju meklēšana attēlos. Trīs metodes, HOG, SIFT un SURF, tiek apskatītas tuvāk, notestētas un salīdzinātas savā starpā. Tā kā HOG metode uzrādīja vislabākos rezultātus, tai tika veikti nopietnāki testi, ar stipri lielāku attēlu kopu. Tika izpētīta iespēja padot HOG metodei dažādas attēla transformācijas, tādējādi ļaujot tai atpazīt arī pagrieztas un citādi transformētas preču zīmes. Iegūtie testēšanas rezultāti uz FlickrLogos-32 attēlu kopas, ļāva salīdzināt metodi ar citām modernajām metodēm. Kopumā HOG metode uzrādīja nedaudz sliktākus rezultātus par citām metodēm. No otras puses, metodes rezultāti joprojām ir konkurētspējīgi, ņemot vērā pielietošanas ērtumu un metodes popularitāti.
This work called “Logo detection in images” presents a study about logo detection in images. Three methods, HOG, SIFT and SURF, were tested and compared to each other. As HOG method showed the best results among the three methods, extra tests with much bigger dataset were performed with this method. In this paper I propose to transform images before launching HOG logo detection that allows to detect even rotated or otherwise transformed logos. Proposed HOG method was tested on FlickrLogos-32 dataset that allowed to compare the results with the state of the art methods. In overall, HOG method produced results that were slightly worse than in other methods. On the other hand, considering popularity and ease of use of the HOG method, it still can be used in many situations.
This work called “Logo detection in images” presents a study about logo detection in images. Three methods, HOG, SIFT and SURF, were tested and compared to each other. As HOG method showed the best results among the three methods, extra tests with much bigger dataset were performed with this method. In this paper I propose to transform images before launching HOG logo detection that allows to detect even rotated or otherwise transformed logos. Proposed HOG method was tested on FlickrLogos-32 dataset that allowed to compare the results with the state of the art methods. In overall, HOG method produced results that were slightly worse than in other methods. On the other hand, considering popularity and ease of use of the HOG method, it still can be used in many situations.