• Journal of IKEEE
• /
• v.22 no.3
• /
• pp.810-815
• /
• 2018

The kernelized correlation filter algorithm yielded meaningful results in accuracy for object tracking. However, because of the use of a fixed size template, we could not cope with the scale change of the tracking object. In this paper, we propose a method to track objects by finding the best scale for each frame using correlation filtering response values in multi-scale using nearest neighbor interpolation and Gaussian normalization. The scale values of the next frame are updated using the optimal scale value of the previous frame and the optimal scale value of the next frame is found again. For the accuracy comparison, the validity of the proposed method is verified by using the VOT2014 data used in the existing kernelized correlation filter algorithm.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.7
• /
• pp.768-774
• /
• 2010

Reconstruction of implicit surfaces from scattered point data sets have been developed in various engineering and scientific studies. In this paper, we represent a method to construct functions of 2D point data using multi-scale kernels and show it can be applied to graphics applications needed to access data in real-time. Our approach is similar to previous work in that a set of coefficients of the functions are calculated and stored in the preprocessing stage and function values at arbitrary positions are evaluated for real-time applications, however, it is different from others in that users can choose detail levels freely in real-time processing stage. The reason why the functions implicitly supports multi-resolution results from the mathematical properties of multi-scale kernels, and proposed method can be expanded to represent multi-resolution functions of n-dimensional data.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.4
• /
• pp.29-37
• /
• 2021

In this paper, we propose a deep learning-based retinal vessel segmentation model for handling multi-scale information of fundus images. we integrate the selective kernel convolution into U-Net-based convolutional neural network. The proposed model extracts and segment features information with various shapes and sizes of retinal blood vessels, which is important information for diagnosing eye-related diseases from fundus images. The proposed model consists of standard convolutions and selective kernel convolutions. While the standard convolutional layer extracts information through the same size kernel size, The selective kernel convolution extracts information from branches with various kernel sizes and combines them by adaptively adjusting them through split-attention. To evaluate the performance of the proposed model, we used the DRIVE and CHASE DB1 datasets and the proposed model showed F1 score of 82.91% and 81.71% on both datasets respectively, confirming that the proposed model is effective in segmenting retinal blood vessels.