An image restoration method for extracting features from three-dimensional auto-stereoscopic integral photography images

Integral Photography (IP) [Hongen Liao 2011] is a visualization technique that produce true three-dimensional images in the real space that can be viewed without supplementary glasses. Using a micro lens array, different aspects of the object can be viewed from different directions, giving observers a sense of depth. Utilizing IP images into augmented reality systems (where images are merged into a real scene) requires an automated alignment between virtual IP images and real objects, leading to the need of understanding the displayed IP images. A natural approach would be capturing multiple views of the IP images and then extracting necessary information from those views. However, since IP images are viewed through a lens array, a lens-like pattern is produced in the observed image(Fig.1(a)). When captured by a camera, simply applying existing image processing methods such as feature extraction would result in unexpected behaviors (e.g. detection of fault features). The purpose of this study is to develop an image restoration method by removing those lens-like pattern from the IP image captured by a camera.

References:

1. Hongen Liao, Takeyoshi Dohi, K. N. 2011. Autostereoscopic 3d display with long visualization depth using referential viewing area based integral photography. IEEE Transactions on Visualization and Computer Graphics 17.
2. Igor Aizenberg, C. B. 2008. A windowed gaussian notch filter for quasi-periodic noise removal. IMAGE AND VISION COMPUTING 26.