Light field image depth estimation method based on hybrid convolutional neural network

Abstract

The invention discloses a light field image depth estimation method based on a hybrid convolutional neural network. The light field image depth estimation method includes the construction of a training data set, the training of a convolutional neural network model, and the generation of a depth estimation map of a light field image. According to the method, the light field depth calculation problem is converted into the classification problem and the relationship between pixel depths in a local area is effectively utilized; and light field data is represented by a four-dimensional parameter. By utilizing the fact that the slope of a straight line in an EPI image of the light field image is proportional to the depth of the scene, the EPI image is used as a medium to map a four-dimensional light field image into a two-dimensional image. By extracting EPI block regions corresponding to pixels in a center-viewing angle image of the light field image, a novel light field image depth estimation training data set is constructed by using a mode of polar line graph block region pairs. According to the light field image depth estimation method, the advantages of deep learning in feature abstraction is utilized, and the accuracy and stability of depth estimation of the method are more advantageous than those of an ordinary conventional methods.

Description

technical field [0001] The present invention relates to the field of light field (Light Field) image and the field of deep learning (Deep Learning), in particular to a method for estimating the depth of light field image based on hybrid convolutional neural network. Background technique [0002] In recent years, with the rapid development of optoelectronic technology, new imaging devices have emerged continuously. As a new three-dimensional imaging technology, light field imaging has attracted the attention of researchers for its unique imaging process. Traditional cameras can only record a two-dimensional plane and cannot obtain scene depth information, while light field imaging can record the four-dimensional position and direction information of light radiation during propagation, so more abundant image information can be obtained during image reconstruction . Light field imaging can realize digital refocusing, synthetic aperture, acquisition of images with large depth o...

AI Technical Summary

Problems solved by technology

[0008] Due to the change of illumination and the complexity of different scenes, the depth estimation performance of existing algorithms is limited to varying degrees, the accuracy of the algorithm needs to be improved, and it is easily affected by noise in the image, and it is easy to be in areas with discontinuous depth and weak texture. Regions are incorrectly estimated, resulting in lower accuracy

Images