Image shadow removal method based on model driving

Abstract

The invention discloses an image shadow removal method based on model driving. The method comprises the following steps: 1, constructing a new adaptive transformation shadow illumination model; 2, constructing a variation model of a shadow-free image according to the shadow illumination model; and 3, constructing an efficient model-driven iterative algorithm to solve the variational model so as to obtain a reconstructed shadow-free image. According to the method, the advantages of interpretability of a traditional model driving scheme and high efficiency of a data driving scheme can be fully combined, an image shadow removing algorithm based on shadow image illumination model driving is constructed, and partial steps in the algorithm are replaced by the convolutional neural network; therefore, shadow removal with high operation efficiency, good removal effect and high interpretability can be realized.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to image shadow removal. An iterative algorithm for removing image shadows based on a shadow image illumination model is proposed, and the proposed iterative optimization algorithm is expanded into the form of a neural network to efficiently complete image shadows. remove. Background technique [0002] The shadow phenomenon is caused by the occlusion of the light source by the object, and it also exists widely in daily scenes. Shadows often bring great challenges to other computer vision tasks, such as object tracking, object detection, etc. Shadow removal is already an important preprocessing step in these computer vision tasks. [0003] There is also a lot of research work in progress in this field. The early traditional shadow removal methods focused on manually designing the relevant priors of shadow images, and building optimized iterative algorithms to o...

AI Technical Summary

Problems solved by technology

But in this way, the neural network often becomes a black box, which lacks interpretability for users, which is also the limitation of this type of method.

[0005] The shadow lighting model in the current mainstream shadow removal methods is based on simplified assumptions, and the design of related physical models is relatively simple. At the same time, deep learning methods lack interpretability, so these methods often produce suboptimal results.

Images