High resolution perspective views extraction based on light field imaging theory
Hui Li, Jing Peng · 2018
The light field imaging system can record the azimuth information of the light. With exposure, the views of different viewpoints can be obtained by calculation. In order to get those views, the traditional algorithm usually uses a periodical method. However, this kind of algorithm ignores the angle-relationship between the viewpoint and the center of each microlens. In this cases, it leads to distortion views. The reconstructed image has a low resolution and poor quality. In order to solve these problems, we studied the principle of light field sampling based on microlens array. According to geometrical optics, we proposed an efficient algorithm to acquire non-periodically the views. Firstly, we calculated the offset of each microlens center by analyzing the optical path of the light field. Then, we analyzed the range of visual angle. We calculated the projection region of the synthetic viewpoint, and extract non-periodically those pixels to form the view. It can improve the resolution of views. Finally, we smoothed and sharpened the result image to further improve the quality of the views, which can preserve the detail information and eliminate the noise at the same time. Experimental results showed that the image quality of the perspective views reconstructed by the proposed algorithm was much better than the conventional method. It had no image superposition and no distortion phenomenon. And the image resolution was much higher and the visual effect was also much better compared to the conventional one.