RGB image spectral reflectance reconstruction method based on fused convolutional neural network

Abstract

The invention provides an RGB image spectral reflectance reconstruction method based on a fused convolutional neural network, and the method comprises the steps: firstly building an image database, and carrying out the batch image preprocessing; training a deep convolution model based on a convolutional neural network by using an image database, and establishing a reconstruction method based on spectral nonlinear mapping and spatial similarity fusion; and finally, acquiring a to-be-reconstructed RGB image, preprocessing the to-be-reconstructed RGB image, inputting the to-be-reconstructed RGB image into the convolutional model for convolutional neural network feature extraction, outputting reconstructed spectral reflectance, and feeding back a reconstruction result to a user to complete a spectral reflectance process. The invention has the beneficial effects that the influence of factors such as image types, image sizes, camera response functions and light source spectral power on the reconstruction result is avoided, the model training time is greatly prolonged, and the problem that the reconstruction precision is not high due to the influence of a linear method or external factors is effectively solved.

Description

technical field [0001] The invention relates to the field of visual spectral reflectance reconstruction, in particular to a method for reconstructing RGB image spectral reflectance based on fusion convolutional neural network. Background technique [0002] In the process of rapid development of digitalization and computer vision, various color printing technologies and image reconstruction technologies have been rapidly popularized, and rich and diverse colors are presented in the eyes of the public, all of which require accurate color representation methods to ensure. Spectral reflectance refers to the ratio of the reflected luminous flux on the surface of an object to the incident luminous flux. It is the essential reason for the different colors of the object and the essential attribute of the surface of the object. It is also the most accurate way to describe the color of the object. Using the spectral reflectance, the object can be obtained under any illumination. and c...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the above-mentioned technical problems of low precision of reflectance reconstruction and large amount of calculation, the present invention provides a method for reconstructing RGB image spectral reflectance based on fusion convolutional neural network. Firstly, an image database is established and the image is preprocessed , secondly, using the powerful image feature extraction ability of the convolutional neural network, a method for reconstructing the spectral reflectance of RGB images based on the fusion convolutional neural network is established. Finally, the spectral reflectance data without spatial structural information output by the spectral CNN unit, Superimposed with the spatial similarity feature information in the RGB image output by the spatial CNN unit to form the final spectral reflectance data

Images