Regularized Discrete Optimal Transport

Abstract

International audienceThis article introduces a generalization of the discrete optimal transport, with applications to color image manipulations. This new formulation includes a relaxation of the mass conservation constraint and a regularization term. These two features are crucial for image processing tasks where one must take into account families of multimodal histograms with large mass variation across modes. The corresponding relaxed and regularized transportation problem is the solution of a convex optimization problem. Depending on the regularization used, this minimization can be solved using standard linear programming methods or first order proximal splitting schemes. The resulting transportation plan can be used as a color transfer map, which is robust to mass variation across image color palettes. Furthermore, the regularization of the transport plan helps remove colorization artifacts due to noise amplification. We also extend this framework to compute the barycenter of distributions. The barycenter is the solution of an optimization problem, which is separately convex with respect to the barycenter and the transportation plans, but not jointly convex. A block coordinate descent scheme converges to a stationary point of the energy. We show that the resulting algorithm can be used for color normalization across several images. The relaxed and regularized barycenter defines a common color palette for those images. Applying color transfer toward this average palette performs a color normalization of the input images

Temporal Upsampling of Point Cloud Sequences by Optimal Transport for Plant Growth Visualization

“…[RTG98, RTG00]. Furthermore, optimal transport has been successfully used for colour transfer in 2D images based on the regularization with the preservation of the gradient of the original picture [PKD07], an approximate variational formulation using an approximate Wasserstein constraint on colour statistics and a generic geometric‐based regularization [RP11], regularized discrete optimal transport within a unified convex variational framework [FPPA14], relaxed optimal transport using a regularization that considers the spatial distribution of colours [RFP14], by using separate illuminant matching and optimal transfer of dominant colours regularized by thin plate splines [FSDH14] and extending standard entropy regularization based optimal transport to unbalanced problems [CPSV16].…”

Wasserstein Loss for Image Synthesis and Restoration

“…We refer to the monograph of Villani for a detailed description of Wasserstein distance [67]. These distances have recently gained some interest in image processing, in particular to perform color and texture manipulation-see for instance [26] and the references therein.…”

Multitemporal classification without new labels: A solution with optimal transport

“…No constraint is applied on the samples in the target domain, as it is supposed to be unlabeled. Note that other forms of regularization have also been proposed in recent literature: for example, authors in [12] proposed a regularization based on the graph Laplacian preserving local relationships during the transport.…”