Light field image super-resolution and depth estimation method and device with cross-scale resolution

Abstract

The invention discloses an optical field image super-resolution and depth estimation method and device based on a trans-scale resolution. The method comprises the following steps that: through a single image super-resolution SISR (Semantic Interpretation For Speech Recognition) method, carrying out upsampling; after a high-resolution image positioned in a middle visual angle is subjected to downsampling, utilizing the SISR method to carry out upsampling, and obtaining an error graph; obtaining a disparity map, utilizing an optical field inherent attribute to carry out global optimization on the disparity map, and meanwhile, carrying out a hole filling operation to enable an area which can not be estimated by the disparity map to be estimated again; utilizing a global disparity map to transfer the error graph to enable the image to recover high-resolution image to realize image super-resolution reconstruction; and according to the super-resolution optical field image, carrying out the deep estimation of a scene to obtain a deep estimation result. By use of the method, the global disparity map can be used for transferring the error graph, so that the image recovers the high-resolution image, the scene depth information is obtained through a high-resolution optical field, in addition, robustness is improved, and calculation speed is high.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method and device for super-resolution and depth estimation of light field images with cross-scale resolution. Background technique [0002] Light field imaging is a brand-new means to record the three-dimensional information of the scene. Different from traditional imaging methods, the light field can not only record the intensity information of light at a certain position, but also record the distribution of light from a certain angle range at this position, so two-dimensional plane imaging is transformed into four-dimensional light field imaging, Includes two spatial dimensions and two angular dimensions. [0003] However, due to the limitation of the sensor in the light field, the spatial and angular resolutions restrict each other. For example, a commercial light field camera, which is composed of multiple micro-lens arrays, is easy to carry, but its spatial resol...

AI Technical Summary

Problems solved by technology

[0003] However, due to the limitation of the sensor in the light field, the spatial and angular resolutions are mutually restricted. For example, a commercial light field camera, which is composed of multiple micro-lens arrays, is easy to carry, but its spatial resolution is very low, which is difficult to meet the demand; One way is a multi-camera array, but this implementation method is expensive, bulky, and needs to be solved

Images