Issn 2091-5446 ilmiy axborotnoma научный вестник scientific journal
Download 1,19 Mb. Pdf ko'rish
|
ilovepdf merged
ILMIY AXBOROTNOMA INFORMATIKA 2021 - yil, 1 - son 111 Table 1. Accuracy (in %) of UzSL gestures when applying SIFT feature extraction method using different training images Classifier No. of Training Images 5 3 1 Min. Dist. 100 99.2 98.9 k-NN (k=5) 100 98.9 98.9 SVM 100 99.2 98.9 In the second scenario, we run an experiment to investigate the effect of changing parameter values of SIFT algorithm. We used four training images and the rest images (i.e. 3 images) are used as testing images. The accuracy of identification is computed when the Peakthr parameter ranged from 0 to 0.2 and different Nangels (2, 4 and 8). Also, we investigated the accuracy when the Psize increased by 4 from 4×4 to reach to 32×32 (i.e we run this experiment 4 times with different parameters). The results of this experiment is shown in Table (2). Table 2. Identification rate (in %) of UzSL based on SIFT feature extraction method using four training images and different values of Peakthr, Psize and Nangels Classifiers PeakThr Psize Nangels 0 0.1 0.2 4x4 8x8 16x16 32x32 2 4 8 NN 100 97.7 94.2 94.2 99.2 100 93.2 94.2 98.9 100 k-NN 100 98.9 96.3 96.3 99.2 100 93.6 96.3 98.9 100 SVM 100 99.2 98.9 97.7 100 100 94.2 96.3 98.9 100 In the third scenario, we run an experiment to prove that our proposed system can overcome the problems of image rotation in different angels. In this scenario, we used SIFT Feature Extraction Methods; and four training images. In the testing phase, the testing images are rotated, then it used to identify the UzSL character. Different orientations are used in our experiment, i.e., the images are rotated in the following angles: (0 ◦ , 45 ◦ , 90 ◦ , 135 ◦ , 180 ◦ , 225 ◦ , 270 ◦ 315 ◦ ). The results of this experiment is shown in Table (3). Download 1,19 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2025
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling