Light field depth estimation method based on frequency domain energy distribution

Abstract

The invention discloses a light field depth estimation method based on frequency domain energy distribution, and the method comprises the steps: extracting a central sub-aperture image of a light field image, and dividing the central sub-aperture image into spatial blocks for frequency domain transformation; for each spatial block, extracting a consistency descriptor and a directivity descriptor from the frequency domain distribution; according to the consistency descriptor of each space block, judging that the space block is a shielded area or a non-shielded area, and applying a judgment result to an angle block corresponding to the space block; for the shielding area, selecting a sub-angle block conforming to consistency description by combining a central angle pixel of the angle block,an edge pixel of the space block and the directivity descriptor; for the non-shielding area, extracting a focusing tensor of the whole angle block; for the shielding area, extracting a focusing tensorof the sub-angle block; establishing a cost equation; and a cost equation is maximized to realize light field depth estimation. According to the method, the depth accuracy of the shielding area is improved while the depth accuracy of the non-shielding area is ensured, so that details at a depth abrupt change position are richer.

Description

technical field [0001] The invention relates to the fields of computer vision and digital image processing, in particular to a method for estimating light field depth based on frequency domain energy distribution. Background technique [0002] The theory of light field imaging has attracted extensive attention from the research and industry circles because of its ability to simultaneously record the spatial information and angle information of light. The light field cameras released based on this theory (such as Lytro, Illum and Raytrix, etc.) Success, not only has strong portability, but also can record four-dimensional light field in one exposure, realize functions such as multi-view conversion, take pictures first and then focus, etc., showing its advantages in the field of computer vision, especially enriching the algorithm of depth estimation in this field, improving In addition to efficiency, it is of great significance for applications such as depth-based 3D reconstru...

AI Technical Summary

Problems solved by technology

However, for all the above algorithms, if the unified focus clue analysis is simply used for all areas, the ambiguity of the front and back of the same viewing angle caused by the occlusion area (that is, the sudden change in depth) will directly cause tensor calculation errors, resulting in most depth estimation algorithms in There is a trend of foreground spreading at the depth mutation, and the result shows a smooth change with glitches, which reduces the accuracy of the algorithm, making the obtained result map poor in quality and low in environmental adaptability

Images