A Feature Point Extraction and Comparison Method Through Representative Frame Extraction and Distortion Correction for 360° Realistic Contents

초록

360° realistic contents are omnidirectional media contents that support front, back, left, right, top and bottom. In addition, they are combined images of images produced using two or more cameras through the stitching process. Therefore 4K UHD is basically supported to represent all directions and distortion occurs in each direction, especially above and below. In this paper, we propose a feature point extraction and similarity comparison method for 360° realistic images by extracting representative frames and correcting distortions. In the proposed method, distortion-less parts for an extracted frame such as the front, back, left, and right directions of the image, except for the largest distortion area such as the up and down directions, are first corrected by a rectangular coordinate system method. Then, the sequence for the similar frames is classified and the representative frame is selected. The feature points are extracted from the selected representative frames by the distortion correction and the similarity can be compared in the subsequent query images. The proposed method is shown, through the experiments, to be superior in speed for the image comparison than other methods, and it is also advantageous when the data to be stored in the server increase in the future. © 2020, Springer Nature Switzerland AG.

키워드